Introduction To Fractional Differential Equations

Nonlinear Systems and Complexity
Series Editor: Albert C. J. Luo
Constantin Milici
Gheorghe Drăgănescu
J. Tenreiro Machado
Introduction
to Fractional
Differential
Equations
www.pdfgrip.com
Volume 25
Series editor
Albert C. J. Luo
Southern Illinois University
Edwardsville, IL, USA
www.pdfgrip.com
Nonlinear Systems and Complexity provides a place to systematically summarize

recent developments, applications, and overall advance in all aspects of nonlinearity,
chaos, and complexity as part of the established research literature, beyond the
novel and recent findings published in primary journals. The aims of the book
series are to publish theories and techniques in nonlinear systems and complexity;
stimulate more research interest on nonlinearity, synchronization, and complexity
in nonlinear science; and fast-scatter the new knowledge to scientists, engineers,
and students in the corresponding fields. Books in this series will focus on the
recent developments, findings and progress on theories, principles, methodology,
computational techniques in nonlinear systems and mathematics with engineering
applications. The Series establishes highly relevant monographs on wide ranging
topics covering fundamental advances and new applications in the field. Topical
areas include, but are not limited to: Nonlinear dynamics Complexity, nonlinearity,
and chaos Computational methods for nonlinear systems Stability, bifurcation,
chaos and fractals in engineering Nonlinear chemical and biological phenomena
Fractional dynamics and applications Discontinuity, synchronization and control.
More information about this series at http://www.springer.com/series/11433

www.pdfgrip.com
Constantin Milici • Gheorghe Drăgănescu

J. Tenreiro Machado
Introduction to Fractional
Differential Equations
123
www.pdfgrip.com
Constantin Milici Gheorghe Drăgănescu

Department of Mathematics Research Center in Theoretical Physics
Polytechnic University of Timişoara West University of Timişoara
Timişoara, Romania Timişoara, Romania
J. Tenreiro Machado
Department of Electrical Engineering
Polytechnic of Porto
Porto, Portugal
ISSN 2195-9994 ISSN 2196-0003 (electronic)

ISBN 978-3-030-00894-9 ISBN 978-3-030-00895-6 (eBook)
https://doi.org/10.1007/978-3-030-00895-6
Library of Congress Control Number: 2018956771
© Springer Nature Switzerland AG 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
www.pdfgrip.com
Preface
The concept of fractional derivative (Fractional derivative (FD)) was introduced

after 1695 as a simply academic generalization of integer derivative. An FD
generalizes the order of differentiation from positive integers Set of natural numbers
(N) to real Set of real numbers (R), or even to complex Set of complex numbers (C)
numbers. A detailed presentation of the old historical steps of fractional calculus
(Fractional calculus (FC)) with references is presented in the papers [45, 50].
New sporadic investigation in this field was published after 1930. A deep
research was carried out by E.L. Post [31, 52]. More recently we find a series of
definitions of the FD [7]. A presentation of the recent history steps can be found in
the papers [46–48, 51]. It was published in a series of monographic books as [6, 21,
22, 26, 37, 38].
In the last two decades, it was established that a series of phenomena can be
studied in terms of FC. It was established that the rheologic properties [34] of some
polymers can be expressed with the aid of fractional differential models [1, 2, 10,
15, 20, 26, 28, 29, 36]. Fractional phenomena were established as the damping
phenomena in the high-density polyurethane foams [42], nuclear reactor dynamics
[35], thermoelasticity [33], mechanical vibrations [8], or biological tissues [5, 19].
An analysis of the integer and fractional entropy is performed in [44].
It was also experimentally verified fractional diffusion phenomena and fractional
electrolytic coating process [4, 30, 32]. This confirms that the roughness of the
electrolytic metallic coatings has a fractal structure and can be described in terms of
fractional diffusion [3] and stochastic differential process [22].
Quantum fractional differential models were studied also in the book of R.
Herrmann [11] or in the papers of Saxena [39], Xiao et al. [55], and Yang et al. [53].
Recently, several fractional devices were developed, containing electrical, ther-
mal, and mechanical components [12, 14]. Also, a series of fractional dynamics
experiments are presented in the book of Biswas et al. [4].
It was designed and has been achieved experimentally a series of fractional
systems [9, 16, 56] involving control [18, 49] and fractional controllers [43]. It was
v
www.pdfgrip.com
vi Preface
introduced a series of identification methods for fractional dynamic systems [13].

Fractional wavelet bases [54] in the field of signal processing were discussed [40].
It is important to tell that FD of a periodic function is also periodic [27].
This review of the possible applications of FC in the real world justifies the
necessity of its extensive study. The aim of this book is to introduce a series of
problems and methods insufficiently discussed in the field of FC.
A series of examples based on symbolic computation, written in Maple© and
Mathematica© , are presented. The reader can find other useful applications for the
case of integer order systems in the book of Inna Shingareva and C. Lizárraga-
Celaya [41], which can be extended to the case of fractional calculus, or the book of
problems [17].
This book is organized in six chapters.
Chapter 1 This chapter presents the most important special functions involved in
FC. A special attention is devoted to the Gamma function, used in FC calculations.
Other special functions such as the Euler, Beta, and Mittag-Leffler functions
(Mittag-Leffler function (MLF)) are also introduced.
Chapter 2 This chapter introduces the fractional integral (Fractional integral (FI))
and FD, in the sense of Riemann–Liouville. The properties of these fractional
operators are discussed.
Chapter 3 This chapter is devoted to the use of the Laplace transform (Laplace
transform (LT)) in FC, because the Riemann–Liouville FI and FD allow the
derivation of closed-form solutions with the aid of the LT method.
Chapter 4 This chapter is devoted to the nonlinear fractional differential equations
(Fractional differential equation (FDE)). The text discusses the generalization of the
methods used to solve the integer order differential equations. In this line of thought,
approaches such as the Picard array, Adomian decomposition, and perturbation
methods are analyzed.
Chapter 5 In this chapter several classical models are generalized in the perspective
of FC. First, the fractional integral sine and cosine functions are formulated, and
the corresponding spirals, projected on the plane, sphere, and cone, are illustrated.
Second, the fractional generalizations of the Lane-Emden, Hermite, Legendre, and
Bessel equations are studied. It is also discussed the power series method for their
solution.
Chapter 6 This chapter extends standard numerical algorithms, namely, the least
squares, Galerkin, and Euler methods, that are applied to FDE. A special attention
is given the Runge–Kutta (Runge–Kutta (RK)) method for fractional equations
and systems. The algorithms are applied to a series of FDE and systems of FDE.
www.pdfgrip.com
Preface vii
Moreover, a generalized method, based on decomposition and LT, is presented [23–

25].
These new algorithms are illustrated by means of the Maple and Mathematica
software packages.
The content of this book is addressed to large category of readers, working in
the fields of fundamental and applied mathematics, theoretical and experimental
physics, experimental engineering, and others. The authors hope that the material
included in the book help researchers to enter to the huge emerging scientific area
of FC and its applications.
Timişoara, Romania Constantin Milici

Timişoara, Romania Gheorghe Drăgănescu
Porto, Portugal J. Tenreiro Machado
References
1. Atanackovič, T. M., Pilipovič, S., Stankovič, B., & Zorica, D. (2014). Frac-
tional calculus with applications in mechanics: Wave propagation, impact and
variational principles. London: Wiley-ISTE.
2. Atanackovič, T. M., Pilipovič, S., Stankovič, B., & Zorica, D. (2014). Frac-
tional calculus with applications in mechanics: Vibrations and diffusion pro-
cesses. London: Wiley-ISTE.
3. Barabási, A.-L., & Stanley, H. E. (2002). Fractal concepts in surface growth
(2nd ed.). Cambridge: Cambridge University Press.
4. Biswas, K., Bohannan, G., Caponetto, R., Mendes Lopes, A., & Tenreiro
Machado, J. A. (2017). Fractional-order devices. SpringerBriefs in nonlinear
circuits. New York: Springer.
5. Bueno-Orovio, A., Kay, D., Grau, V., Rodriguez, B., & Burrage, K. (2014).
Fractional diffusion models of cardiac electrical propagation: Role of structural
heterogeneity in dispersion of repolarization. Journal of the Royal Society
Interface, 11, 20140352.
6. Das, S. (2011). Functional fractional calculus. Berlin: Springer.
7. de Oliveira, E. C., & Tenreiro Machado, J. A. (2014). A review of definitions
for fractional derivatives and integral. Mathematical Problems in Engineering,
2014, Article ID 238459.
8. Deü, J.-F., & Matignon, D. (2010). Simulation of fractionally damped mechan-
ical systems by means of a Newmark-diffusive scheme. Computers & Mathe-
matics with Applications, 59, 1745–1753.
9. Diethelm, K., Ford, N. J., & Freed, A. D. (2002). Simulation of fractionally
damped mechanical systems by means of a Newmark-diffusive scheme. Non-
linear Dynamics, 59, 3–22.
www.pdfgrip.com
viii Preface
10. Drăgănescu, G. E. (2006). Application of a variational iteration method to

linear and nonlinear viscoelastic models with fractional derivatives. Journal of
Mathematical Physics, 47, 082902.
11. Herrmann, R. (2011). Fractional calculus. An introduction for physicists.
Singapore: World Scientific Publishing Company.
12. Hughes, R. N. (1992). Design and analysis of electrical circuits that pro-
duce fractional-order differentiation. Master’s thesis, Air University, Wright-
Patterson Air Force Base, Ohio. AFIT/GE/ENG/92M-05.
13. Jalloul, A., Jelassi, K., & Trigeassou, J.-C. (2013). A comparative study
of identification techniques for fractional models. International Journal of
Electrical & Computer Engineering, 3, 186–196.
14. Jesus, I. S., Tenreiro Machado, J. A., & Boaventura Cunha, J. (2008). Fractional
electrical impedances in botanical elements. Journal of Vibration and Control,
14, 1389–1402.
15. Kilbas, A. A., Srivastava, H. M., & Trujillo, J. J. (2006). Theory and applica-
tions of fractional differential equations. Amsterdam: Elsevier.
16. Klafter, J., Lim, S. C., & Metzler, R. (2011). Fractional dynamics: Recent
advances. Singapore: World Scientific Publishing Company.
17. Krasnov, M. L., Kiselyov, A. I., & Makarenko, G. I. (1981). A book of problems
in ordinary differential equations. Moscow: Mir Publishers.
18. Luo, Y., & Chen, Y. (2012). Fractional order motion controls. Chichester: John
Wiley & Sons.
19. Magin, R. L. (2010). Fractional calculus models of complex dynamics in
biological tissuese. Computers & Mathematics with Applications, 59, 1586–
1593.
20. Mainardi, F., & Spada, G. (2011). Creep, relaxation and viscosity properties for
basic fractional models in rheology. The European Physical Journal Special
Topics, 193, 133–160.
21. Malinowska, A. B., Odzijewicz, T., & Torres, D. F. M. (2015). Advanced
methods in the fractional calculus of variations. SpringerBriefs in applied
sciences and technology. Heidelberg: Springer.
22. Meerschaert, M. M., & Sikorskii, A. (2012). Stochastic models for fractional
calculus. De Gruyter studies in mathematics. Heidelberg: Walter de Gruyter
GmbH & Co.
23. Milici, C., & Drăgănescu, G. (2014). A method for solve the nonlinear
fractional differential equations. Saarbrücken: Lambert Academic Publishing.
24. Milici, C., & Drăgănescu, G. (2015). New methods and problems in fractional
calculus. Saarbrücken: Lambert Academic Publishing.
25. Milici, C., & Drăgănescu, G. (2017). Introduction to fractional calculus.
Saarbrücken: Lambert Academic Publishing.
26. Oldham, K. B., & Spanier, J. (1973). The fractional calculus: Theory and
applications of differentiation and integration to arbitrary order. San Diego:
Academic Press.
www.pdfgrip.com
Preface ix
27. Ortigueira, M. D., Tenreiro Machado, J., & Trujillo, J. J. (2017). Fractional
derivatives and periodic functions. International Journal of Dynamics and
Control, 5, 72–78.
28. Oustaloup, A. (2014). Diversity and non-integer differentiation for system
dynamics. Hoboken: Wiley-ISTE.
29. Petráš, I. (2011). Fractional-order nonlinear systems modeling, analysis and
simulation. Beijing/Heidelberg: Higher Education Press/Springer.
30. Podlubny, I. (1998). Fractional differential equations: An introduction to frac-
tional derivatives, fractional differential equations, to methods of their solution
and some of their applications. Mathematics in science and engineering. San
Diego: Academic Press.
31. Post, E. L. (1930). Generalized differentiation. Transactions of the American
Mathematical Society, 32, 723–781.
32. Povstenko, Y. (2015). Linear fractional diffusion-wave equation for scientists
and engineers. Cham: Birkhäuser.
33. Povstenko, Y. (2015). Fractional thermoelasticity. Cham: Springer.
34. Pritchard, R. H., & Terentjev, E. M. (2017). Oscillations and damping in the
fractional Maxwell materials. Journal of Rheology, 61, 187–203.
35. Ray, S. S. (2016). Fractional calculus with applications for nuclear reactor
dynamics. Boca Raton: CRC Press.
36. Rossikhin, Yu. A., & Shitikova, M. V. (2004). Analysis of the viscoelastic
rod dynamics via models involving fractional derivatives or operators of two
different orders. The Shock and Vibration Digest, 36, 3–26.
37. Sabatier, J., Agrawal, O. P., & Tenreiro Machado, J. A. (Eds.). (2007). Advances
in fractional calculus theoretical developments and applications in physics and
engineering. Dordrecht: Springer.
38. Samko, S. G., Kilbas, A. A., & Marichev, O. I. (Eds.). (1993). Fractional
integrals and derivatives: Theory and applications. Yverdon: Gordon and
Breach Science Publishers.
39. Saxena, R. K., Saxena, R., & Kalla, S. L. (2010). Solution of spacetime
fractional Schrödinger equation occurring in quantum mechanics. Fractional
Calculus & Applied Analysis, 13, 177–190.
40. Sheng, H., Chen, Y. & Qiu, T. (Eds.). (2012). Fractional processes and
fractional-order signal processing: Techniques and applications. London:
Springer.
41. Shingareva, I. K., & Lizárraga-Celaya, C. (Eds.). (2011). Solving nonlinear
partial differential equations with maple and mathematica. Wien: Springer.
42. Singh, R., Davies, P., & Bajaj, A. K. (2003). Identification of nonlinear and
viscoelastic properties of flexible polyurethane foam. Nonlinear Dynamics, 34,
319–346.
43. Tenreiro Machado, J. (2001). Discrete-time fractional-order controllers. Frac-
tional Calculus & Applied Analysis, 4, 47–66.
44. Tenreiro Machado, J. A. (2010). Entropy analysis of integer and fractional
dynamical systems. Nonlinear Dynamics, 62, 371–378.
www.pdfgrip.com
x Preface
45. Tenreiro Machado, J. A., Mainardi, F., & Kiryakova, V. (2010). A poster about
the old history of fractional calculus. Fractional Calculus & Applied Analysis,
13, 447–454.
46. Tenreiro Machado, J., Kiryakova, V., & Mainardi, F. (2011). Recent history
of fractional calculus. Communications in Nonlinear Science and Numerical
Simulation, 16, 1140–1153.
47. Tenreiro Machado, J., Lopes, A. M., Duarte, F. B., Ortigueira, M. D., & Rato,
R. T. (2014). Rapsody in fractional. Fractional Calculus & Applied Analysis,
17, 1188–1214.
48. Tenreiro Machado, J. A., Mainardi, F., Kiryakova, V., & Atanacković, T. (2016).
Fractional calculus: D’où venons-nous? Que sommes-nous? Où allons-nous?
Fractional Calculus & Applied Analysis, 17, 1074–1104.
49. Valério, D., & Costa, J. S. (2013). An introduction to fractional control.
London: CRC Press.
50. Valério, D., Tenreiro Machado, J. A., & Kiryakova, V. (2014). Some pioneers of
the applications of fractional calculus. Fractional Calculus & Applied Analysis,
17, 552–578.
51. Weilbeer, M. (2005). Efficient numerical methods for fractional differential
equations and their analytical background. PhD thesis, Facultät für Mathematik
und Informatik, Technischen Univerisität Braunschweig, Braunschweig.
52. Wheeler, N. (1997). Construction and physical application of fractional calcu-
lus. Technical report, Reed College Physics Department.
53. Yang, X.-J., Baleanu, D., & Tenreiro Machado, J. A. (2013). Mathematical
aspects of the Heisenberg uncertainty principle within local fractional Fourier
analysis. Boundary Value Problems, 131, 1–16.
54. Yang, X.-J., Baleanu, D., Srivastava, H. M., & Tenreiro Machado, J. A.
(2013). On local fractional continuous wavelet transform. Abstract and Applied
Analysis, 2013, 5.
55. Zhang, X., Wei, C., Liu, Y., & Luo, M. (2014). Fractional corresponding oper-
ator in quantum mechanics and applications: A uniform fractional Schrödinger
equation in form and fractional quantization methods. Annals of Physics, 350,
124–136.
56. Zhou, Y., Ionescu, C., & Tenreiro Machado, J. A. (2015). Fractional dynamics
and its applications. Nonlinear Dynamics, 80, 1661–1664.
www.pdfgrip.com
Contents
1 Special Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Euler’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Gamma Function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Beta Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Integral Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Mittag-Leffler Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Function E(t, α, a). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 Fractional Derivative and Fractional Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1 Fractional Integral and Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3 The Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1 Calculus of the Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Calculus of the Original Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.1 Calculus of Original Using Residues . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.2 Calculus of Original with Post’s Inversion Formula . . . . . . . . . . 36
3.3 The Properties of the Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.1 The Property of Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.2 Similarity Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.3 The Differentiation and Integration Theorems . . . . . . . . . . . . . . . . 37
3.3.4 Delay Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3.5 Displacement Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3.6 Multiplication Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3.7 Properties of the Inverse Laplace Transform . . . . . . . . . . . . . . . . . . 39
3.4 Laplace Transform of the Fractional Integrals and Derivatives . . . . . . . 43
3.4.1 Fractional Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.2 Fractional Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
xi
www.pdfgrip.com
xii Contents
4 Fractional Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.1 The Existence and Uniqueness Theorem for Initial Value
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Linear Fractional Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 Nonlinear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 The Adomian Decomposition Method. . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.2 Decomposition of Nonlinear Equations . . . . . . . . . . . . . . . . . . . . . . . 67
4.3.3 Perturbation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4 Fractional Systems of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4.1 Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4.2 Nonlinear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5 Generalized Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.1 Cornu Fractional System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.1.1 Cos and Sin Fractional of Type Fresnel . . . . . . . . . . . . . . . . . . . . . . . 87
5.1.2 Cornu Fractional System and Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.1.3 Cornu Generalized Curve/System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.1.4 Cornu Fractional System in a Plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.1.5 Fractional Cornu Spiral on the Sphere . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.1.6 Fractional Cornu Spiral on the Cone. . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.2.1 The Müntz Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.2.2 Lane-Emden Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.2.3 The Taylor Series Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.2.4 The Generalized Hermite Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.2.5 The Generalized Legendre Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.2.6 The Generalized Bessel Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.2.7 Nonlinear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6 Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.1 Variational Iteration Method for Fractional Differential Equations . . 121
6.2 The Least Squares Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.3 The Galerkin Method for Fractional Differential Equations . . . . . . . . . . 135
6.4 Euler’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.5 Runge–Kutta Methods for Fractional Differential Equation . . . . . . . . . . 149
6.5.1 The Second Order Runge–Kutta Method . . . . . . . . . . . . . . . . . . . . . . 150
6.5.2 The Fourth Order Runge–Kutta Method. . . . . . . . . . . . . . . . . . . . . . . 153
6.5.3 A More General System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.5.4 A Vectorial Runge–Kutta Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 179
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
www.pdfgrip.com
Acronyms
FC Fractional calculus
FD Fractional derivative
FDE Fractional differential equation
FI Fractional integral
LT Laplace transform
RK Runge–Kutta
RK2 Second order Runge–Kutta
RK4 Fourth order Runge–Kutta
VIM Variational iteration method
List of Symbols
E(t, α, a) E function
Jp (t) Bessel function of first kind
B Beta function
erfc Complementary error function
C Set of complex numbers
det Determinant
erf Error function
γ Euler’s constant
Ei Exponential integral function
Gamma function
erfi Imaginary error function
L Laplace transform
max Maximum
min Minimum
Eα One parameter Mittag-Leffler function
N Set of natural numbers
R Set of real numbers
Re Real part
Res Residues
sup Supremum
Eα,β Two parameter Mittag-Leffler function
www.pdfgrip.com
Chapter 1
Special Functions
In this chapter several special functions used in the follow-up of the book are
presented briefly. More details about these functions can be found in [1, 3, 4, 8].
1.1 Euler’s Function
1.1.1 Gamma Function
We start by considering the Gamma function, or second order Euler1 integral,

denoted (·) represented in Fig. 1.1.
Function Gamma function () is defined as:
∞
(p) = e−x x p−1 dx. (1.1)
0
Theorem Function (p) is convergent for p > 0.

Proof The integral can be written as:
1 ∞
(p) = e−x x p−1 dx + e−x x p−1 dx = I1 + I2 ,
0 1
1
where I1 = e−x x p−1 dx is convergent.
0
1 L. Euler (1707–1783).
© Springer Nature Switzerland AG 2019 1

C. Milici et al., Introduction to Fractional Differential Equations, Nonlinear
Systems and Complexity 25, https://doi.org/10.1007/978-3-030-00895-6_1
www.pdfgrip.com
2 1 Special Functions
Fig. 1.1 The plot of

y = (x) function 6
4
y
–4 –3 –2 –1 0 1 2 3 4
x
–2
–4
–6
Since e−x is decreasing on the interval [0, 1], from x = 0, we have:

1 1 1
e−x x p−1 dx < x p−1 dx = .
0 0 p
∞
Moreover, I2 = e−x x p−1 dx is also convergent. We obtain:
1
x p−1
1 ≤ x ⇒ x p−1 e−x ≤ e−x/2 ⇔ x p−1 ≤ ex/2 ⇔ ≤ 1.
ex/2
x p−1
Because lim = 0, we have:
x→∞ ex/2
∞ ∞
e−x x p−1 dx ≤ e−x/2 dx = 2e−1/2 .
1 1
The integral (1.1) is convergent for p > 0 and divergent for p ≤ 0.

The basic properties of the Gamma function are:
1. The function (p) is continuous for p > 0.
2. The function (p) obeys the property:
(p + 1) = p (p). (1.2)

www.pdfgrip.com
1.1 Euler’s Function 3
∞ ∞
−x p

Proof (p+1) = e x dx = − e−x x p ]∞
0 +p e−x x p−1 dx = p(p).
0 0
3. The following relations are also valid:
(p + n) = (p + n − 1) . . . (p + 1) p (p), (1.3)
(1) = 1,
(n + 1) = n!,
(0) = +∞.
4. For p = −n it results:
(−n + 1)
(−n) =
−n
(−n + 2) (−n + 3) (0)
= = = . . . = (−1)n = (−1)n ∞.
n(n − 1) n(n − 1)(n − 2) n!
(p + 1)
5. Taking account that the function can be written as (p) = , it
p
results that the function can be defined also for negative values of p, in the
interval −1 < p < 0.
If −n < p < −(n − 1), then from (1.3) it results:
(p + n)
(p) = .
p(p + 1) . . . (p + n − 1)
Using the substitution p + n = α, it results after calculations:
(−1)n (α)
(α − n) = .
(1 − α)(2 − α) . . . (n − α)
6. Using the identity (1.2) we obtain:

1 1 1 1
m+ = 1+ m− = m− m−
2 2 2 2

1 3 1
= m− m− m−
2 2 2

1 3 531 1
= m− m− ... ,
2 2 222 2
www.pdfgrip.com
or:

1 (2m − 1)! 1 (2m)! 1
m+ = = .
2 2m 2 m!22m 2
7. We can prove the identity [4]:

1 1
p−1
(p) = ln dy
0 y
8. The following particular values for function can be useful for calculation
purposes:

1 √
= π,
2

1 √
− = −2 π ,
2

3 1 1 1 1√
= 1+ = = π,
2 2 2 2 2

4! 1
5 1 2 3√
= 2+ = 4
= π,
2 2 2! 2 4

1 1 1 4 . . . (3m − 2) 1
= 2.678938 m+ = ,
3 3 3m 3

2 2 2 5 . . . (3m − 1) 2
= 1.354118 m+ = ,
3 3 3m 3

1 1 1 5 . . . (4m − 3) 1
= 3.625600 m+ = ,
4 4 4m 4

3 3 3 7 . . . (4m − 1) 3
= 1.225417 m+ = .
4 4 4m 4
9.
(p + 1) p
= ,
(q + 1)(p − q + 1) q
www.pdfgrip.com
10. The Gauss’2 formula is:

∞
1
1 p p −1
(p) = 1+ 1+ .
p k k
k=1
Proof We express e−x as:

x k
e−x = lim 1− .
k→∞ k
Then, we obtain:

∞
−x p−1
k x k p−1
(p) = e x dx = lim 1− x dx.
0 k→∞ 0 k
For x = t k ⇒ dx = k dt, resulting:

1
(p) = lim k p
(1 − t)k t p−1 dt.
k→∞ 0
Integrating by parts we obtain:

1 1 1
1 1 1 k 1
(1 − t) dt =
k p
(1 − t) t
k p
− t d(1 − t) =
p k
(1 − t)k−1 t p dt.
p 0 p 0 p 0 p 0
Repeating this operation it follows:

k p k!
(p) = lim .
k→∞ p(p + 1) . . . (p + k)
But:
(k + 1)p
lim = 1.
k→∞ kp
From:
1 1 2p 3p . . . (k + 1)p
(p) = lim ,
p k→∞ (1 + p)(1 + p/2) . . . (1 + p/k) 1p 2p . . . k p
it follows:
∞
1 p p −1
∞
1
1 (k + 1)p 1
(p) = = 1 + 1+ ,
p (1 + p/k) k p p k k
k=1 k=1
2 J.C.F. Gauss (1777–1855).

www.pdfgrip.com
excepting the values Re(p) = 0, −1, −2, . . ..

11. The Weierstrass3 form of the Gamma function is:
∞

1 1 −p/k
= pe γp
1+ e
(p) k
k=1
valid, excepting the values Re(p) = 0, −1, −2, . . . , −n, . . . , where Real
part (Re) represents the real part and n ∈ N.
Here the symbol Euler’s constant (γ ) represents the Euler’s constant, given by:
p
1
γ = lim − ln p = 0.577215663 . . .
p→∞ k
k=1
Proof The Gauss’ formula can be written as:

k p k!
(p) = lim ,
k→∞ p(p + 1) . . . (p + k)
excepting for Re(p) = 0, −1, −2, . . .. We can write also:
k p = ep ln k = ep(ln k − 1 − 1/2 − . . . − 1/k) ep + p/2 + . . . + p/k ,
ep(ln k − 1 − 1/2 − . . . − 1/k) ep + p/2 + . . . + p/k

k (t) = ,
p(1 + p)(1 + p/2) . . . (1 + p/k)
p p p
1 1 e e2 ek
k (t) = exp p ln k − 1 − − . . . − ... ,
2 k 1 + p 1 + p2 1 + pk
and finally:
1 1 p p
p
= lim = epγ lim e−p (1 + p)e− 2 . . . e− k 1 +
(p) k→∞ k (t) k→∞ k
∞

p −p
= peγp 1+ e k.
k
k=1
12. Reflection property of the Gamma function is given by:

π
(p) (1 − p) = , 0 ≤ p ≤ 1.
sin(πp)
3 K.T.W. Weierstrass (1815–1897).

www.pdfgrip.com
Proof From the Weierstrass’ formula we have:

∞
1 1
p −p/k p p/k
= −p2 eγp e−γp 1+ e 1− e
(p) (−p) k k
k=1
∞

p2
= −p2 1− .
k2
k=1
(1 − p)
But (−p) = . It results:
p
∞

1 1 p2
=p 1− 2 .
(p) (1 − p) k
k=1
∞

p2
Because sin(nπ ) = π n 1 − 2 it results the reflection property.
k
k=1
Another Proof
∞ ∞
−t p−1
(p)(1 − p) = e t dt e−s s 1−p−1 ds
0 0
∞ ∞ p
−(t+s) t
= e t −1 dtds.
0 0 s
Using the changes of variables

t
t + s = u, = v,
s
and the residue theory it can be obtained:
1 p
v π
(p)(1 − p) = dv = .
0 1+v sin(pπ )
1
Figure 1.2 depicts (x) .
1.1.2 Beta Function
Here we consider the Beta function, denoted Beta function (B).

The Beta function, or the first order Euler function, can be defined as [2, 7]:
1
B(p, q) = x p−1 (1 − x)q−1 dx,
0
where Re(p) > 0 and Re(q) > 0.

www.pdfgrip.com
Fig. 1.2 The plot of

y = 1/ (x) function 2
y
1
–4 –3 –2 –1 0 1 2 3
x
–1
–2
In the following we will enumerate the basic properties of the Beta function:
1. For every p > 0 and q > 0, we have:
B(p, q) = B(q, p).
2. For every p > 0 and q > 1, the Beta function B satisfies the property:
q −1
B(p, q) = B(p, q − 1).
p+q −1
Proof
1
B(p, q) = x p−1 (1 − x)q−1 dx,
0
x p (1 − x)q−2 = x p−1 (1 − x)q−2 − x p−1 (1 − x)q−1 ,

1 xp x p (1 − x)q−1 1 q − 1 1 p
B(p, q) = (1 − x) d =
q−1
+ x (1 − x)q−2 dx
0 p p 0 p 0

q − 1 1 p−1 q − 1 1 p−1
= x (1 − x) dx −
q−2
x (1 − x)q−1 dx
p 0 p 0
q −1 q −1
= B(p, q − 1) − B(p, q).
p p
www.pdfgrip.com
3. For every p > 0 and q > 0, it is valid the identity:
(p)(q)
B(p, q) = .
(p + q)
Proof The product (p)(q) can be written as:

∞ ∞ ∞ ∞
−t p−1 −s p−1
(p)(q) = e t dt e s ds = e−(t+s) t p−1 s q−1 dtds,
0 0 0 0
∞ ∞
(p + q) = e−(t+s) t p−1 s q−1 dt ds.
0 0
We use the notation t + s = x, for 0 < t < ∞ and 0 < t < ∞.

The Jacobian is

y x
D[t, s]
=
= −xy − x + xy = −x,

D[x, y]
1 − y −x
resulting:

D[t, s]
dtds = dxdy = xdxdy,
D[x, y]
1 1
(p)(q) = e−x (xy)p−1 x q−1 (1 − y)q−1 xdxdy
0 0
∞ 1
−x p+q−1
= e x dx y p−1 (1 − y)q−1 dy,
0 0
(p)(q) = (p + q)B(p, q).
4. For every p > 0, and for the natural number n, it can be proved
1 · 2 · 3 . . . (n − 1)
B(p, n) = B(n, p) = ,
p(p + 1) . . . (p + n)
and also:
1
B(p, 1) = .
p
www.pdfgrip.com
For any natural numbers m, n we obtain:
(n − 1)!(m − 1)!
B(m, n) = .
(m + n − 1)!
5. Legendre4 duplication formula:

22p−1 1
(2p) = √ (p) p + .
π 2
Proof Substituting p = q in B(p, q):

1
(p)(q)
B(p, q) = = up−1 (1 − u)q−1 du,
γ (p + q) 0
1+x
and replacing u = , we have:
2
p−1 p−1
2 (p) 1 1 1+x 1−x 1 1 p−1
= = 1 − x2 dx,
(2p) 2 −1 2 2 22p−1 −1
resulting
1
22p−1 2 (p) = 2(2p) (1 − x 2 )p−1 dx.
0

1 1 1 √
But B ,p = 2 (1 − x )
2 p−1
dx, = π , resulting finally:
2 0 2

1
(p)
1 2
22p−1 2 (p) = (2p)B , p = (2p) .
2 1
+p
2
6. Triplication formula:
1
33p− 2 1 2
(3p) = √ (p) p + p+ .
π 2 3
4 A.M. Legendre (1752–1833).

www.pdfgrip.com
1.3 Mittag-Leffler Function 11
7. Gauss multiplication formula:

1 k−1
(p) p + ... p + = (2k)(k−1)/2 k −kp+1 (kp).
k k
1.2 Integral Functions
In this section we introduce the error, imaginary error, complementary error,

and exponential integral functions, denoted as erf(·), erfi(·), erfc(·), and Ei(·),
respectively.
Function Error function (erf) is defined as:
z
2a
e−a z dz.
2 2
erf(az) = √
π 0
Function Imaginary error function (erfi) is defined as:

z
2a 2 2
erfi(az) = √ ea z dz = −i erf(iaz)
π 0
Function Complementary error function (erfc) is defined as:

∞
2a
e−a z dz
2 2
erfc(az) = √
π z
Function Exponential integral function (Ei) is defined as:

z az
e
Ei(az) = dz.
−∞ z
1.3 Mittag-Leffler Function
In this section we introduce the one- and two-parameter Mittag-Leffler functions,

denoted as Eα (·) and Eα,β (·), respectively.
The one-parameter Mittag-Leffler5 function, One parameter Mittag-Leffler func-
tion (Eα ), is defined as:
5 M.G. Mittag-Leffler (1846–1927).

www.pdfgrip.com
∞
zk
Eα (z) = Re(α) > 0.
(αk + 1)
k=0
The two-parameter Mittag-Leffler function, Two parameter Mittag-Leffler func-

tion (Eα,β ), is defined as:
∞
zk
Eα,β (z) = Re(α), Re(β) > 0, β ∈ C.
(αk + β)
k=0
For particular values of α and β it results:

∞ ∞

zk
E0,1 (z) = = zk , (1.4)
(1)
k=0 k=0
∞
∞
zk
zk
E1,1 (z) = = = ez , (1.5)
(k + 1) k!
k=0 k=0
∞
zk∞ ∞
zk 1 zk+1 ez − 1
E1,2 (z) = = = = , (1.6)
(k + 2) (k + 1)! z (k + 1)! z
k=0 k=0 k=0
∞
∞ ∞
zk zk 1 zk+2 ez − 1 − z
E1,3 (z) = = = 2 = ,
(k + 3) (k + 2)! z (k + 2)! z2
k=0 k=0 k=0
(1.7)
∞
zk
E1,0 (z) = = ze ,
z
(1.8)
(k)
k=0
eat √
E1, 3 (at) = √ erf( at), (1.9)
2 at
1 √ √
E1, 1 (at) = √ + eat at erf( at), (1.10)
2 π

1 eat √ 2
E1, 5 (at) = √ erf( at) − √ , (1.11)
2 at at π

1 1 √ √
E1,− 1 (at) = − √ + (at) √ + at eat erf( at) , (1.12)
2 2 π π
1
Eα,β (t) = + tEα,α+β (t), (1.13)
(β)
www.pdfgrip.com
1.4 Function E(t, α, a) 13
∞
∞
∞

tk t k+1 tk
Eα,β (t) = = =
(αk + β) (α(k + 1) + β) (αk + α + β)
k=0 k=−1 k=−1
1
= + tEα,α+β (t),
(β)
(k + m)! ∞
(m) dm tk
Eα,β (z) = Eα,β (z) = . (1.14)
dzm k! (αk + αm + β)
k=0
1.4 Function E(t, α, a)
The function E function (E(t, α, a)) is defined as:

∞
(at)k
E(t, α, a) = = t α E1,α+1 (at), (1.15)
(k + α + 1)
k=0
or in the integral form:

t
1
E(t, α, a) = τ α−1 ea(t−τ ) dτ. (1.16)
(α) 0
We prove schematically the equivalence between the forms (1.15) and (1.16). We
denote the integral:
t
1
I = τ α−1 ea(t−τ ) dτ
(α) 0
∞
1 t a k (t − τ )k
= τ α−1
dτ
(α) 0 k!
k=0
∞
1 ak t k t τ k
= τ α−1
1− dτ .
(α) k! 0 t
k=0
τ
For u = we obtain:
t
1
t τ k
I1 = τ α−1 1 − dτ = t α−1 uα−1 (1 − u)k du
0 t 0
www.pdfgrip.com
1 t α (α)(k + 1)
= tα uα−1 (1 − u)k du = t α B(α, k + 1) = .
0 (α + k + 1)
It results:
∞
(at)k
I = tα = t α E1,α+1 (at) = E(t, α, a).
(k + α + 1)
k=0
In (1.16) we take α = 1
2 and u2 = aτ . It results after calculations:
eat √
E1, 3 (at) = √ erf(a t).
2 at
Using the definition we obtain:
∞
(at)k
E1, 1 (at) = .
2
k=0 k + 12
Replacing k with k + 1 we have:

⎡ ⎤
∞
∞

(at)k+1 (at)−1 (at)k
E1, 1 (at) = = (at) ⎣ + ⎦
2
k=−1 k + 1
2 1
2 k=0 k + 3
2
1 1 √ √
= √ + atE1, 3 (at) = √ + ateat erf( at)
π 2 π
results finally (1.10).

In order to obtain (1.11), the parameter k will be replaced by k − 1 in E1, 5 (at).
2
To obtain (1.12) the parameter k will be replaced by k + 2 in E1,− 1 (at).
2
For further details about the Mittag-Leffler function, readers can check [5].
There are known a series of fractional trigonometric functions, useful to solve
FDE, studied in detail in [6].
References
1. Abramowitz, M., & Stegun, I. A. (Eds.). (1965). Handbook of mathematical functions. Dover
books on mathematics. New York: Dover Publications.
2. Brychkov, Y. A. (2008). Handbook of special functions derivatives, integrals, series and other
formulas. Boca Raton: Chapman and Hall/CRC.
3. Erdélyi, A. (1954). Tables of integral transforms (Vols. 1–2). New York: McGraw-Hill Book
Company.
4. Erdélyi, A. (Ed.). (1955). Higher transcendental functions (Vols. I–III). New York: McGraw-
Hill Book Company
www.pdfgrip.com
References 15
5. Gorenflo, R., Kilbas, A. A., Mainardi, F., & Rogosin, S. V. (2014). Mittag-Leffler functions,
related topics and applications. Springer monographs in mathematics. Heidelberg: Springer.
6. Lorenzo, C. F., & Hartley, T. T. (2017). The fractional trigonometry: With applications to
fractional differential equations and science. Hoboken: Wiley & Sons, Inc.
7. Nikiforov, A. F., & Ouvarov, V. (1976). Eléments de la théorie des fonctions spéciales. Moscow:
Mir Publishers.
8. Podlubny, I. (1998). Fractional differential equations: An introduction to fractional derivatives,
fractional differential equations, to methods of their solution and some of their applications.
Mathematics in science and engineering. San Diego: Academic Press.
www.pdfgrip.com
Chapter 2
Fractional Derivative and Fractional
Integral
2.1 Fractional Integral and Derivative
Definition (Fractional Integral of Order α) For every α > 0 and a local inte-
grable function f (t), the right FI of order α is defined:
t
1
α
a It f (t) = (t − u)α−1 f (u)du, −∞ ≤ a < t < ∞. (2.1)
(α) a
Alternatively, it can be defined also the left FI as:

b
1
α
t Ib f (t) = (u − t)α−1 f (u)du, −∞ < t < b ≤ ∞. (2.2)
(α) t
For particular values of the a and b parameters, the following cases are known:
• Riemann1 : a = 0, b = +∞
• Liouville2 : a = −∞, b = 0.
Definition (Fractional Derivative of Order α) For every α, and n = α
the
Riemann–Liouville derivative of order α can be defined as:
1 d n t
a Dt f (t) = (t − u)n−α−1 f (u) du.
α
(2.3)
(n − α) dx a
1 G.F.B. Riemann (1826–1866).

2 J. Liouville (1809–1882).

www.pdfgrip.com
18 2 Fractional Derivative and Fractional Integral
Theorem 1 The following integration rules are valid:

b b
φ(x)a Ixα ψ(x)dx = ψ(x)x Ibα φ(x) dx, (2.4)
a a
b b
f (x)a Dxα g(x)dx = g(x)x Dbα f (x) dx. (2.5)
a a
Also, it must be noticed that:

α α
a Ix a Dx f (x) = f (x),
where 0 < α < 1.

Proof We use the Dirichlet3 theorem, written in the form:
b x b b
dx f (x, t)dt = dt f (x, t) dx.
a a a t
– For the case (2.4), we introduce the following notation:
1 φ(x)ψ(t)
f (x, t) = ,
(α) (x − t)1−α
in the Dirichlet theorem. It results:

b x b b
1 ψ(t) 1 φ(t)
φ(x)dx dt = ψ(t)dt dt,
a (α) a (x − t)1−α a (α) t (x − t)1−α
b b
φ(x)a Ixα ψ(x)dx = ψ(x)x Ibα φ(x) dx
a a
– For case (2.5), we introduce in (2.4):

α
x Db f (x) = φ(x), α
a Dx g(x) = ψ(x), α α
a Ix a Dx f (x) = f (x).
Theorem 2 The following integration and derivation rules are valid:

α+1 (t − a)α
(a) a It [Df (t)] = a Itα f (t) − f (a),
(α + 1)
3 J.P.G.L. Dirichlet (1805–1859).

www.pdfgrip.com
2.1 Fractional Integral and Derivative 19

n (t − a)α−k

a It [a Dt f (t)] = f (t) −
α−k
(b) α α D
a t f (t) ,
t=a (α − k + 1)
k=1
(t − a)α−1
(c) D[a Itα f (t)] = a Itα [Df (t)] + f (a),
(α)
α+p
D k f (a)(t − a)α+k
p−1
(d) a Itα f (t) = a It [D p f (t)] + , where p is a positive
(α + k + 1)
k=0
integer.
D k f (a)(t − a)α+k
p−1
(e) D p [a Itα f (t)] = a Itα [D p f (t)] + , where p is positive
(α + k + 1)
k=0
integer.
Proof
(a) Integrating by parts, it results:
t t
1 1
a It
α+1
[Df (t)] = (t − u)α f (u)du = [(t − u)α f (u)]
(α + 1) a (α + 1) a
t
α
+ (t − u)α−1 f (u) du,
(α + 1) a

(t − a)α α t
a It
α+1
[Df (t)] = − f (a) + (t − u)α−1 f (u) du.
(α + 1) α(α) a
(b) This formula can be verified by induction, using (a), or:

t
1
I = a Itα [a Dtα f (t)] = (t − u)α−1 [a Duα f (u)] du,
(α) a
t
1 d
I = (t − u)α [a Duα f (u)] du,
(α + 1) 0 dt
t
(t − a)α 1 d
I =− a D α
f (t) + (t − u)α+1 [a Duα−1 f (u)] du.
(α + 1) t t=a (α + 2) 0 dt
(t − a)α (t − a)α+1 α−1

I =− α
a Dt f (t) − a Dt f (t)
(α + 1) t=a (α + 2) t=a
t
1 d
+ (t − u)α+1 [a Duα−1 f (u)] du,
(α + 2) 0 dt
...

n (t − a)α−k

I = f (t) − α−k
a Dt f (t) .
t=a (α − k + 1)
k=1
www.pdfgrip.com
(c) In the FI, we have:

t
1
a It f (t) = (t − u)α−1 f (u) du,
α
(α) a
and we make the change of variable: u = t − x 1/α (see also [3]). We obtain:
(t−a)α
1
α
a It f (t) = f (t − x 1/α ) dx.
(1 + α) a
Then, for t > 0:

(t−a)α
1 ∂
D[a Itα f (t)] = α(t − a) f (a) +
α−1
f (t − x ) dx .
1/α
(1 + α) a ∂t
Reversing the change of variable t − x 1/α = u, we obtain:

t
1 α−1 ∂
D[a Itα f (t)] = α(t − a) f (a) + α
α−1
(t − u) f (u) du .
(1 + α) a ∂t
Hence:
(t − a)α−1 f (a)
D[a Itα f (t)] = + a Itα [Df (t)].
(α)
(d) Replacing α by α + 1 and f by Df in (a), we have:
(t − a)α f (a)
a It
α+1
[D 2 f (t)] = a Itα+1 [Df (t)] − .
(α + 1)
Replacing a Itα+1 [Df (t)] with (a), we obtain:
(t − a)α f (a) Df (a) (t − a)α−1

a It
α+1
[D 2 f (t)] = a Itα [f (t)] − − .
(α + 1) (α + 1)
(d) Can be established by repeated iterations.

(e) For t > 0, we must differentiate (c):
Df (a) (t − a)α−2
D 2 [a Itα f (t)] = D[a Itα f (t)] + .
(α − 1)
[Df (a) (t − a)α−2 ] [Df (a) (t − a)α−1 ]
D 2 [a Itα f (t)] = a Itα f (t) + + ,
(α − 1) (α)
and by repeated iterations we obtain (e).

www.pdfgrip.com
Theorem 3 The exponents property:
β α+β
α
a It a It f (t) = a It f (t).
For this theorem we recommend also [4].

Proof For α > 0, β > 0, it results:
t u
β 1 1
I = a Itα a It f (t) = (t − u)α−1 (u − v)β−1 f (v) du dv.
(α) (β) a a
If we apply the Dirichlet equality

t u t t
f (v)dudv = f (v) du dv,
a a a v
we obtain:
t t
1 1
I= (t − u)α−1 (u − v)β−1 f (v) dudv,
(α) (β) a v
u = v + z(t − v),
du = (t − v)dz, t − u = (1 − z)(t − v),

t 1
1 1
I= (t − v) α+β−1
f (v) (1 − z)α−1 zβ−1 dv dz,
(α) (β) a 0
but:
1 (α)(β)
(1 − z)α−1 zβ−1 du = B(α − 1, β − 1) = .
0 (α + β)
Finally, it results:
t
1 α+β
I= (t − v)α+β−1 f (v)dv = a It f (t).
(α + β) a
Theorem 4

β α−β
(a) a Dtα a It f (t) = a Dt f (t).
m
(t − a)α−k
β α−β β−k
(b) a It a Dt f (t) = a It f (t) −
α
a Dt f (t) .
(α + 1 − k) t=a
k=1
where m = β
+ 1.
www.pdfgrip.com

m (t − a)−α−k
β α+β β−k
(c) α
a Dt D
a t f (t) = D
a t f (t) − D
a t f (t) .
t=a (1 − α − k)
k=1
Proof
(a)
d n n−α β d n n−(α−β)
β α−β
α
a Dt a It f (t) = a It a It f (t) = a It = a Dt .
dt n dt n
(b)
t
β 1 β
I = a Itα a Dt f (t) = (t − u)α−1 a Dt f (u) du
(α) a
t
1 d β
I= (t − u)α a Dt f (u) du
(α + 1) a dt
t
(t − a)α β 1 d
β−1

I= − D
a t f (t) + (t − u)α+1 a Dt f (u) du
(α+1) t=a (α+2) a dt
...
Example Solve the following FDE with initial value:
D 1/2 y(t) = y(t),

√
D −1/2 y(0) = −2 π ,
transforming it in a first order differential equation.

Solution Using the theorem 4 (c), we obtain:
t −1/2−1
D 1/2 D 1/2 y(t) = y (t) − D 1/2−1 y(0) = D 1/2 y(t) = y(t),
(1 − 1/2 − 1)
y (t) − t −3/2 = y(t).
Theorem 5 Linearity property:
a It [C1 f (t) + C2 g(t)] = C1 a Itα f (t) + C2 a Itα g(t),

α
where: C1 and C2 are constants and f (t) and g(t) are two arbitrary functions.
www.pdfgrip.com
Proof
t
1
a Dt [C1 f (t) + C2 g(t)] = (t − y)α−1 [C1 f (y) + C2 g(y)] dy
α
(α) a
t
1
= C1 (t − y)α−1 f (y)dy
(α) a
t
1
+C2 (t − y)α−1 g(y)dy
(α) a
= C1 a Itα f (t) + C2 a Itα g(t).
Theorem 6 If the function f (t) possess continuous derivative, then for α >
0, n = α
+ 1:

n−1 (k) t
f (t − a)k−α 1
α
a It f (t) = + (t − y)n−α−1 f (n) (y) dy.
(k + 1 − α) (n − α) a
k=0
Proof In the Cauchy4 formula:

t
1
a It f (t) = (t − u)α−1 f (u)du,
α
(α) a
it will be applied successively the integration by parts formula (see [5]).

Theorem 7 We denote 0 Itα with I α , for p ∈ N, α > 0. It can be proved that:
p k
−α d
(a) I α [t p f (t)] = t p I α+k f (t),
k dt k
k=0
p k
α d
(b) D α [t p f (t)] = t p D α−k f (t).
k dt k
k=0
Proof
t
1
(a) I = I α [t p f (t)] = (t − u)α−1 up f (u) du,
(α) 0

p
(−1)k p!t p−k
u = (t − (t − u)) =
p p
(t − u)k ,
k!(p − k)!
k=0
4 A.L. Cauchy (1789–1857).

www.pdfgrip.com

p t
p! 1 1
I= (−1)k t p−k (t − u)α+k−1 f (u) du,
(p − k)! (α) 0 k!
k=0

p
(α + k − 1) . . . α d k p α+k
I= (−1)k t I f (t),
k! dt k
k=0

p α d k
I= (−1)k t p I α+k f (t).
k dt k
k=0
d n n−α p
(b) I = D α [t p f (t)] = I [t f (t)]. We obtain:
dt n
p
−n+α n d k+j
n
I= t p D α−k−j f (t)
k j dt k=j
k=0 j =0
i
−n+α n
p
d i p n−i
I= i
t D f (t) ,
dt i−j j
i=0 j =0
but
i
−n+α n α
= ,
i−j j i
j =0
thus:
p k
α d
D α [t p f (t)] = t p D α−k f (t).
k dt k
k=0
Definition of Caputo Fractional Derivative

Let α > 0, n = α
. The Caputo5 derivative operator of order α is defined as [1, 2]:
t d n
1
C α
a Dt f (t) = (t − u)n−α−1 f (u) du. (2.6)
(n − α) a du
For a = 0, we introduce the notation:

C
Dtα f (t) = D α f (t).
5 M. Caputo (1967–).
www.pdfgrip.com
Theorem 8 For t > 0, α ∈ R, n − 1 < α < n, n ∈ N, and a function f (t) which

obey the conditions of Taylor6 theorem, the following representation is valid:

n−1
f (k) (a)
a Dt f (t) = a Dt f (t) + (t − a)k−α .
α C α
(1 + k − α)
k=0
Proof In order to simplify our presentation, we consider a = 0.

Because f (t) can be expanded in Taylor series we can write

n−1
tk
f (t) = f (k) (0) + Rn−1 ,
(k + 1)
k=0
where:

t f (n) (y)(t − y)n−1 1 t
Rn−1 = dy = f (n) (y)(t − y)n−1 dy = I n f (n) .
0 (n − 1)! (n) 0
If we apply the operator D α we obtain successively:

n−1 n−1
tk Dα t k
α
D f (t)=D α
f (0) + Rn−1 =
(k)
f (k) (0) + D α Rn−1 ,
(k+1) (k + 1)
k=0 k=0

n−1
t k−α
D α f (t) = f (k) (0) + I n−α f n (t),
(k − α + 1)
k=0

n−1
t k−α
D α f (t) = f (k) (0) + D α f (t).
(k − α + 1)
k=0
The Caputo Fractional Derivative in the Origin

For a function f (t), for which f (t) = 0, if t < 0, it can be defined:
t
1
C α
0 Dt f (t) = (t − u)n−α−1 f (n) (u) du,
(n − α) 0
where α
= n.
Observation
If C is a constant, then:
C α
0 Dt C = 0,
6 B. Taylor (1685–1731).
www.pdfgrip.com
and the Riemann–Liouville FD of C is:
C x −α
α
0 Dt C = , α = 1, 2, . . . .
(1 − α)
In what follows we note the Caputo derivative in the origin, simply, using the
notation D α f (x).
Theorem 9 If n − 1 < α < n, where n ∈ N, and α ∈ R, then:
lim D α f (t) = f (n) (t),

α→n
lim D α f (t) = f (n−1) (t) − f (n−1) (0).

α→n−1
Proof In the formula

t
1 f (n) (y)dy
D f (t) =
α
,
(n − α) 0 (t − y)α+1−n
we will use the integration by parts, obtaining:

t t t

u(y)v (y) dy = u(y) v(y) − u (y) v(y) dy,
0 0 0
u(y) = f (n) (y), v (y) = (t − y)n−α−1 ,
u (y) = f (n+1) (y), v(y) = −(t − y)n−α .
It results:

1 (t − y)n−α t
D α f (t) = −f (n) (y)
(n − α) n−α 0
t
1
+ (t − y) n−α (n+1)
f (y)dy .
n−α 0
Using the property of function
(n − α + 1) = (n − α)(n − α),
it results:
t
1
D α f (t) = f (n) (0) + f (n+1) (y)(t − y)n−α dy ,
(n − α + 1) 0
www.pdfgrip.com
t t

lim D f (t) = f
α (n)
(0) + f (n+1) (y)dy ] = f (n) (0) + f (n) (y) = f (n) (t),
α→n 0 0
t
lim D α f (t) = f (n) (0) + f (n+1) (y)(t − y)dy
α→n−1 0
t

= f (n) (0)t + (t − y)f (n) (y) = f (n−1) (t) − f (n−1) (0).
0
Example 1 Let us calculate the FD for α > 0, n − 1 < α < n, β > n − 1 of the
function f (t) = t β using the definitions, for the case:
1. Riemann–Liouville.
2. Caputo in the origin, using the definition.
Solution
1. For the Riemann–Liouville derivative, we can write:
t
1 dn
I =D t = α β
uβ (t − u)n−α−1 du,
(n − α) dt n 0
and we take:
u = v t, du = t dv.
It follows:
t
1 dn
I = (vt)β [(1 − v)t]n−α−1 t dv
(n − α) dt n 0
t
1 dn
= (1 − v)n−α−1 v β t n−α+β dv,
(n − α) dt n 0
t
1 d n n−α+β
I= (1 − v)n−α−1 v β t dv,
(n − α) 0 dt n
but
dn λ (λ + 1) λ−n
t = t ,
dt n (λ − n + 1)
1
B(p, q) = v p−1 (1 − v)q−1 dv,
0
www.pdfgrip.com
so that it results:

1 (n − α + β + 1) −α+β 1
I= t (1 − v)n−α−1 v β dv,
(n − α) (−α + β + 1) 0
1 (n − α)(β + 1)
(1 − v)n−α−1 v β dv = B(n − α, β + 1) = ,
0 (n − α + β + 1)
(β + 1)
Dα t β = I = t β−α .
(−α + β + 1)
2. In this case we apply the definition of the Caputo derivative of t β :

t
1 (uβ )(n)
I =D t =α β
du,
(n − α) 0 (t − u)α+1−β

1 t (β + 1)
I= uβ−n (t − u)n−α−1 du.
(n − α) 0 (β − n + 1)
We use the change of variable u = vt, resulting after calculations:
du = t dv,

(β + 1) 1
I= (uv)β−n [(t − v)n−α−1 ]t dv.
(n − α)(β − n + 1) 0
Finally, we obtain:
(β + 1) (β + 1) β−α

I= B(β − n + 1, n − α) = t .
(n − α)(β − n + 1) (β − α + 1)
Example 2 Find the Riemann–Liouville FI and FD of
f (t) = (t − a)β .
Solution For the FI we apply the Riemann–Liouville definition:

t
1
I= α
a It f (t) = (t − u)α−1 (u − a)β du.
(α) a
The following change of variable
u−a
= v,
t −a
du = (t − a)dv,
www.pdfgrip.com
allows to calculate:

(t − a)α+β 1 (t − a)α+β
I= (1 − v)α−1 v β dv = B(α, β + 1),
(α) 0 (α)
(β + 1)
I= (t − a)α+β .
(α + β + 1)
For the FD we apply the Riemann–Liouville definition:

d n n−α
Df = a Dtα (t − a)β = aI (t − a)β ,
dt n
and finally:
(β + 1) dn (β + 1)
Df = (t − a)β+n−α = (t − a)β−α .
(β + n − α + 1) dt n (β − α + 1)
Theorem 10 If n − 1 < α < n and if f (t) satisfy the conditions of the Taylor
theorem, then:
∞
f (k) (0)
D f (t) =
α
t k−α .
(k − α + 1)
k=0
Proof Because f (t) satisfy the conditions of the Taylor theorem, we can apply the
Taylor expansion:
∞ ∞
f (k) (0) k f (k) (0) k
f (t) = t = t .
k! (k + 1)
k=0 k=0
The FD will be
∞
∞
f (k) (0) α k f (k) (0) (k + 1) k−α
D α f (t) = D t = t ,
(k + 1) (k + 1) (k − α + 1)
k=0 k=0
and finally:
∞
f (k) (0)
D α f (t) = t k−α .
(k − α + 1)
k=0
Theorem 11 The integration and derivation rules are valid:

r k
−α d
I [t f (t)] =
α r
t r I α+k f (t), r ∈ N, α > 0
k dt k
k=0
r k
α d
D α [t r f (t)] = t r D α−k f (t), r ∈ N, α ∈ R.
k dt k
k−0
www.pdfgrip.com
Proof Using the definition

t
1
I α [t r f (t)] = (t − τ )α−1 τ r f (τ )dτ
(α) 0
and taking

r
(−1)k r! r−k
τ = [t − (t − τ )] =
r r
t (t − r)k ,
k!(r − k)!
k=0
we obtain:

r t
(−1)k r! r−k 1 1
I α [t r f (t)] = t (t − τ )α+k−1 f (τ )dτ.
k!(r − k)! (α) 0 k!
k=0
But:
t
1
I α+k
f (t) = (t − τ )α+k−1 f (τ )dτ,
(α + k) 0
(α + k) = (α + k − 1)(α + k − 1) = . . . = (α + k − 1) . . . α(α).
Finally, we obtain:

r
(α + k − 1) . . . α d k r α+k
I α [t r f (t)] = (−1)k t I f (t)
k! dt k
k=0
r k
−α d
= t r I α+k f (t).
k dt k
k=0
Observation
For 0 < α < 1, f (0) = 0, we have:
t
1
I α f (t) = (t − u)α−1 f (u) du,
(α) 0
t t
α 1
I α f (t) = (t − u)α−1 f (u) du = f (u) (du)α .
α(α) 0 (α + 1) 0
www.pdfgrip.com
References 31
References
1. Caputo, M. (1999). Lessons on seismology and rheological tectonics. Technical report, Univer-
sitá degli Studi La Sapienza, Rome.
2. Ishteva, M., Scherer, R., & Boyadjiev, L. (2003). On the Caputo operator of fractional calculus
and C-Laguerre functions. Technical report, Bulgarian Ministry of Education and Science, Grant
MM 1305/2003.
3. Miller, K. S., & Ross, B. (1993). An introduction to the fractional calculus and fractional
differential equations. New York: John Wiley & Sons.
4. Valério, D., & Costa, J. S. (2013). An introduction to fractional control. London: CRC Press.
5. Podlubny, I. (1998). Fractional differential equations: An introduction to fractional derivatives,
fractional differential equations, to methods of their solution and some of their applications.
Mathematics in science and engineering. San Diego: Academic Press.
www.pdfgrip.com
Chapter 3
The Laplace Transform
A function f (t) is called original function if [8, 9]:

1. f (t) ≡ 0 for t < 0,
2. |f (t)| < Mes0 t for t > 0 with M > 0, s0 ∈ R.
3. For every closed interval [a, b], the function satisfies the Dirichlet conditions:
(a) is bounded,
(b) or is continuous, or has a finite number of discontinuities of first kind,
(c) has a finite number of extremes.
We consider the complex variable s = α + iβ, where Re(s) = α ≥ s1 ≥ s0 .
Then
∞
F (s) = e−st f (t) dt, (3.1)
0
is called the Laplace1 integral, or Laplace transform (LT), or image of the original
function f (t). In the follow-up we denote by L[f (t)] = F (s) or simply by Laplace
transform (L) the Laplace transform. In Table 3.1 the LT of some elementary usual
functions are listed.
The corresponding inverse Laplace transform is [1, 4]:
γ +it
1
f (t) = lim F (s)est dt = L−1 [F (s)], (3.2)
2π i t→∞ γ −it
√
where i = −1 and γ ∈ R, so that the contour path of integration is contained in
the convergence region.
1 Pierre Laplace (1749–1827).

www.pdfgrip.com
34 3 The Laplace Transform
Table 3.1 Images of basic elementary functions

Number Original Image Number Original Image
1 1 1 7 eαt cos βt s−α
s (s − α)2 + β 2
2 tn 1 8 eαt sin βt β
n! s n+1 (s − α)2 + β 2
3 eαt 1 9 t n αt 1
e
s−α n! (s − α)n+1
4 cos βt s 10 t cos βt s2 − β2
s2 + β2 (s 2 + β 2 )2
5 sin βt β 11 t sin βt 2sβ
s2 + β2 (s 2 + β 2 )2
6 cosh(βt) s 12 sinh(βt) β
s2 − β2 s2 − β2
3.1 Calculus of the Images
Example 1 Establish the image of f (t) = t λ :

∞
F (s) = e−ts t λ dt.
0
We introduce the change of variable x = ts. We have also dx = s dt. It results:

∞
x λ dx
F (s) = e−x λ ,
0 s s
∞
1 (λ + 1)
F (s) = λ+1 e−x x λ dx = .
s 0 s λ+1
The direct and inverse LT are:

(λ + 1) −1 1 tλ
L(t ) =
λ
, L = .
s λ+1 s λ+1 (λ + 1)
Example 2 Find the image of: f (t) = sin2 (t).

1 − cos(2t)
Using the identity sin2 t = , it results:
2
1 2s 2
F (s) = − = .
2s 2(s 2 + 4) s(s 2 + 4)
1 bt 1 −bt
Example 3 Find the image of f (t) = te + te .
2 2
1 1 s 2 + b2
F (s) = + = .
2(s − b)2 2(s + b)2 (s 2 − b2 )2
www.pdfgrip.com
3.2 Calculus of the Original Function 35
3.2 Calculus of the Original Function
3.2.1 Calculus of Original Using Residues
If we denote by L[f (t)] = F (s), then if we consider all residues of the function
F (s)est , denoted by r1 , r2 , . . . rn we can use the theorem:
f (t) = r1 + r2 + . . . + rn .
The residues, denoted by Residues (Res), can be calculated using the following
procedure (theorem):
If a is a simple pole of the function, then:
Res[est F (s)] = lim [(s − a)est F (s)].

a s→a
If a is a simple pole of order n of the function, then:
1
Res[est F (s)] = lim [(s − a)n est F (s)](n−1) .
a (n − 1)! s→a
Example 1 Find the original function of the image:
1
F (s) = .
(s − 3)2 (s + 1)
Solution The residue of the function F (s)est is
est e−t
r1 = lim (s + 1) = ,
t→−1 (s − 3)2 (s + 1) 16
st
d e
r2 = lim
t→3 ds s+1
te3t − e3t
= ,
16
resulting:
e−t te3t − e3t

f (t) = r1 + r2 = + .
16 16
Example 2 Find the original function of the following image:
s2
F (s) = .
s4 −1
www.pdfgrip.com
Solution The function F has singularities: 1, −1, −i, i. The residues will be:
s 2 est et
r1 = Res F (s)est = lim (s − 1) = ,
1 s→1 (s − 1)(s + 1)(s 2 + 1) 4
s 2 est e−t
r2 = Res F (s)est = lim (s + 1) = − ,
−1 s→−1 (s − 1)(s + 1)(s 2 + 1) 4
s 2 est −e−it
r3 = Res F (s)est = lim (s + i) = ,
−i s→−i (s 2 − 1)(s − i)(s + i) 4i
s 2 est eit
r4 = Res F (s)est = lim (s + i) = .
i s→i (s 2 − 1)(s − i)(s + i) 4i
It results finally

1 et − e−t 1 eit − e−it
f (t) = + ,
2 2 2 2i
or:
1
f (t) = (sinh t + sin t).
2
3.2.2 Calculus of Original with Post’s Inversion Formula
E. Post2 obtained the formula [7, 10]:

k+1
(−1)k k k
f (t) = lim F (k) , t > 0.
k→∞ k! t t
Example 3 Find the original function of the image:
n!
F (s) = n!s −1−n .
s n+1
Solution With the aid of Post formula we have

−n−k−1
k k+1 (n + k)! k
f (t) = t lim
n
.
k→∞ k!t k+1 t
2 E. Post (1897–1954).
www.pdfgrip.com
3.3 The Properties of the Laplace Transform 37
Using the Stirling3 formula:
k!
lim √ k k e−k = 1,
k→∞ 2π k
it results:

n −n n n k n n
f (t) = t e lim 1+ 1+ 1+ = t n.
k→∞ k k k
3.3 The Properties of the Laplace Transform
In this section, we will use the notations F (s) = L[f (t)] and L(s) = G[g(t)]. In
the follow-up are discussed properties of the LT.
3.3.1 The Property of Linearity
L[af (t) + bg(t)] = aF (s) + bG(s), a, b ∈ R. (3.3)
3.3.2 Similarity Theorem

1 s
L[f (αt)] = F , α > 0. (3.4)
α α
3.3.3 The Differentiation and Integration Theorems
Theorem (Differentiation of an Original) The LT of the derivative of order k from

f (t) gives:

L[f (k) (t)] = s k F (s) − s k−1 f (0) + s k−2 f (0) + . . . + f (k−1) (0) . (3.5)
Proof For k = 1, using the definition and integrating by parts, we have:

∞ ∞ ∞

L[f (t)] = e−st f (t)dt = e−st f (t) + s e−st f (t) dt
0 0 0
= −f (0) + sL[f (t)],
3 J. Stirling (1692–1770).
www.pdfgrip.com
L[f (t)] = sF (s) − f (0),
and for k = 2, using L[f (t)] = L[(f (t)) ] we obtain:
L[f (t)] = s 2 F (s) − sf (0) − f (0).
Using mathematical induction method, we have: L[f (k) (t)].

Example 1 Find the LT for original f (t) = t 2 .
Solution
f (t) = 2t, f (t) = 2,
f (0) = 0, f (0) = 0, f (0) = 2,
L[f (0)] = s 2 F (s) − sf (0) − f (0),
2 2
L[2] = = s 2 F (s) ⇒ F (s) = 3 .
s s
Example 2 Find the LT of following original f (t) = cos 2t.
Solution f (t) = −2 sin(2t), f (t) = −4 cos(2t),
f (0) = 1, f (0) = 0, f (0) = −4,
L[f (t)] = s 2 F (s) − sf (0) − f (0), ⇒ −4F (s) = s 2 F (s) − s,
s
F (s) = .
s2 + 4
Theorem (Integration of an Original) It can be obtained:

t
F (s)
L f (τ )dτ = ,
0 s
t
Proof Let be g(t) = f (τ )dτ . Then:
0
L[g (t)] = sL[g] − g(0).
Also, we have g (t) = f (t) and g(0) = 0. Then:
F (s)
L[g] = .
s
Theorem (Differentiation of a Transform) We have:
F (n) (s) = L[(−t)n f (t)].

www.pdfgrip.com
Proof The theorem can be proved by induction. For n = 1, we have, successively:
F (s) = L[−tf (t)],

∞ ∞
dF (s) d
= e−st f (t)dt = − e−st tf (t)dt = −L[tf (t)],
ds ds 0 0
and, finally:
d
F (n) (s) = [F (n−1) (s)].
ds
3.3.4 Delay Theorem
For a positive number a we have:
L[f (t − a)] = e−as F (s)
3.3.5 Displacement Theorem
It is valid the formula:
L[eλt f (t)] = F (s − λ).
3.3.6 Multiplication Theorem
The convolution product of two functions f (t) and g(t) is designated by the symbol
∗. We have:
t
(f ∗ g)(t) = f (τ )g(t − τ ) dτ,
0
L[(f ∗ g)(t)] = F (s)G(s), F (s) = L[f (t)], G(s) = L[g(t)].
3.3.7 Properties of the Inverse Laplace Transform
The following formula is valid:

The inverse LT is not unique. We have:

−1 s −(α−β)
L = t α−1 Eβ,α (at β ), α, β > 0, s α > |a|, (3.6)
sβ − a
www.pdfgrip.com

−1 s −(α−1)
L = t α−1 E1,α (at) = E(t, α − 1, a), (3.7)
s−a

−1 s −α
L = tE(t, α, a) − αE(t, α + 1, a), (3.8)
(s − a)2

−1 s −α 1 α(α + 1)
L = t 2 E(t, α, a) − αtE(t, α + 1, a) + E(t, α + 2, a),
(s − a) 3 2 2
(3.9)
n+k
∞ (−a) k
1
L−1 = t α(n+1)−1 k
t k(α−β) ,
(s α + as β )n+1 [k(α − β) + (n + 1)α]
k=0
(3.10)
where 0 < β ≤ α.
n+k
∞
∞ (−b)n (−a)k
sγ
L−1 = t α−γ −1 k
t k(α−β)+nα ,
s α + as β + b [k(α − β) + (n + 1)α − γ ]
n=0 k=0
(3.11)
where β ≤ α, γ < α, a ∈ R or: |a| < s α−β , |b| < |s α + as β |.

Proof Proof of the identity (3.6):
∞
L t α−1 Eβ,α (at β ) = e−st t α−1 Eβ,α (at β )dt
0
∞
∞
ak
= e−st t kβ+α−1 dt
(kβ + α) 0
k=0
∞
ak
= L[t kβ+α−1 ]
(kβ + α)
k=0
1 a k
∞
∞
ak (kβ + α) s −(α−β)
= = = .
(kβ + α) s kβ+α sα sβ sβ − a
k=0 k=0
Proof of the identity (3.8):
d
L[t E(t, α, a) − αE(t, α + 1, a)] = − L[E(t, α, a)] − αL[E(t, α + 1, a)]
ds
−α
d s s −(α+1) 1
=− −α = α .
ds s − a s−a s (s − a)2
www.pdfgrip.com

1 2 1
L t E(t, α, a) − αtE(t, α + 1, a) + α(α + 1)E(t, α + 2, a)
2 2
1 d2 d α(α + 1)
= 2
L [E(t, α, a)] + α L [E(t, α + 1, a)] + L [E(t, α + 2, a)]
2 ds ds 2
−α
1 d2 s d s −(α+1) α(α + 1) s −(α+2) 1
= +α + = α .
2 ds 2 s−a ds s−a 2 s−a s (s − a)3
For the identities (3.10) and (3.11) the reader can use the reference [5].
Proof of the identity (3.10). We will apply the well-known identity [4]:
1 n+k
∞
= (−x)k .
(1 + x) n k
k=0
It follows:
1 1 1
= α n+1
(s α + as β )n+1 (s ) a n+1
1 + αβ
s
1 n+k −a k
∞
= .
(s α )n+1 k s α−β
k=0

∞

sγ sγ 1 s γ (−b)n
= = ,
s α + as β + b s α + as β b (s α + as β )n+1
1+ n=0
sα + as β
and for the case of (3.10) we obtain:
sγ sγ 1
=
(s + s a)
α β n+1 s α(n+1) a n+1
1 + α−β
s
1 n+k −a k
∞ ∞
n+k (−a)k
= α(n+1)−γ = ,
s k s α−β k s α(n+1)+k(α−β)−γ
k=0 k=0
sγ
∞
n+k
∞
(−a)k
= (−b) n
,
s α + as β + b k s α(n+1)+k(α−β)−γ
n=0 k=0
www.pdfgrip.com
n+k
∞
∞ (−b)n (−a)k
sγ
L−1 = t α−γ −1 k
t k(α−β)+nα .
s α + as β + b [k(α − β) + (n + 1)α − γ ]
n=0 k=0
Lemma The following identities are valid:

n+k
∞
∞ (−b)n (−a)k
1
L−1 = t α−1 k
t k(α−1)+nα ,
s α + as + b [k(α − 1) + (n + 1)α]
n=0 k=0
(3.12)
for 1 ≤ α, 0 < α, a ∈ R, and |a| < s α−1 , |b| < |s α + as|, respectively:
n+k
∞
∞ (−b)n (−a)k
s α−1
L−1 = k
t k(α−1)+nα , (3.13)
s α + as + b [k(α − 1) + nα + 1]
n=0 k=0
for 1 ≤ α, a ∈ R and for |a| < s α−1 , |b| < |s α + as|.
Proof Proof of the identity (3.12). In (3.11) we take γ = 0, β = 1.

Proof of the identity (3.13). In (3.11) we take γ = α − 1, and β = 1.
Example 1 Establish the LT of:
f (t) = y (t) − 2y (t) − 3y(t); where: y(0) = y (0) = 0.
Solution
F (s) = s 2 Y (s) − sy(0) − y (0) − 2[sY (s) − y(0)] − 3Y (s),
and finally:
F (s) = (s 2 − 2s − 3)Y (s).

t
y= ydt + 1.
0
Solution
Y (s) 1 1
Y (s) = + ⇒ Y (s) = .
s s s−1

t
y(τ ) sin(t − τ )dτ = 1 − cos t.
0
www.pdfgrip.com
3.4 Laplace Transform of the Fractional Integrals and Derivatives 43
Solution
1 1 s 1 1
Y (s) = − 2 = , ⇒ Y (s) = .
s2 + 1 s s +1 s(s 2 + 1) s

t
y(τ )et−τ dτ = y(t) − et .
0
Solution
1 1 1
Y (s) = Y (s) − ; ⇒ Y (s) = .
s−1 s−1 s−2
3.4 Laplace Transform of the Fractional Integrals

and Derivatives
3.4.1 Fractional Integrals
If α > 0, the Riemann–Liouville and Caputo FI are the same for both cases:
t
1
I = I f (t) =
α
(t − y)α−1 f (y)dy.
(α) 0
Using the LT of the convolution product formula, we have:
1 F (s)
L[I ] = L[t α−1 ]L[f (t)] = α .
(α) s
3.4.2 Fractional Derivatives
– The Riemann–Liouville FD is
n t
1 dt
L Dtα f (t) = L (t − u) n−α−1
f (u)du
(n − α) d n t 0
n
dt
=L I n−α f (t) ,
d nt
where we can apply the classical formula:
L[f (n) (t)] = s n F (s) − s n−1 f (0) − . . . − f (n−1) (0),

www.pdfgrip.com
F (s) n−α−1 k n−α

n−1
L[Dtα f (t)] = s n − s [D I f (t)]t=0 ,
s n−α
k=0

n−1
L[Dtα f (t)] = s α F (s) − s n−α−1 [D k I n−α f (t)]t=0 .
k=0
– The Caputo [2, 3] FD is

t
1
L[D f (t)] = L
α
(t − u) n−α−1 (n)
f (u)du = L[I n−α f (n) (t)],
(n − α) 0
where we can apply the classical formula:
L[f (n) (t)] = s n F (s) − s n−1 f (0) − . . . − f (n−1) (0),
and
F (s) n−α−1 n−α (k)

n−1
L[D α f (t)] = s n − s [I f (t)]t=0 .
s n−α
k=0

n−1
L[D α f (t)] = s α F (s) − s n−α−1 f (k) (0).
k=0
Exercise 5 For the function f (t) = t 2 , calculate the Caputo L[D α ]. It results:
1
1. α = ,
2
1
2. α = − .
2
Solution
1.
1
L[D 1/2 t 2 ] = L[2t],
1− 21
s
2
L[D 1/2 t 2 ] = 5
,
s2
3 3
2 2t 2 8t 2
D 1/2 t 2 = L−1 5
= = √ .
s2 52 3 π
www.pdfgrip.com
References 45
In MAPLE, the FD of order 1/2 of t from t 2 can be evaluated using the

command: fracdiff(tˆ 2,t,1/2).
For 0 < α < 1, f α (0) = 0, and
t t
F (t) = f (u)(du) = α
α
(t − u)α−1 f (u) du
0 0
we define:
∞
Lα [f (t)] = Fα (s) = Eα (−s α t α )f (t) (dt)α .
0
The following formulae can be obtained without difficulty [6]:

1. Lα [t α f (t)] = −D α Lα [f (t)].
1
2. Lα [f (at)] = α Lα [f (t)].
a
3. Lα [f
(tt − b)] = Eα (−s b )Lα [f (t)].
α α
1
4. Lα f (u) (du)α = α Lα [f (t)].
0 a (α + 1)
5. Lα [g (α) (t)] = s α Lα [g(t)] − (α + 1)g(0).
For
t
(f (t) ∗ g(t))α = f (t − u)g(u)(du)α ,
0
we have:

Lα (f (t) ∗ g(t))α = Lα [f (t)] Lα [g(t)].
References
1. Abramowitz, M., & Stegun, I. A. (Eds.). (1965). Handbook of mathematical functions. Dover
books on mathematics. New York: Dover Publications.
2. Caputo, M. (1999). Lessons on seismology and rheological tectonics. Technical report,
Universitá degli Studi La Sapienza, Rome.
3. Caputo, M., & Fabrizio, M. (2015). Damage and fatigue described by a fractional derivative
model. Journal of Computational Physics, 293, 400–408.
4. Debnath, L., & Bhatta, D. (2007). Integral transforms and their applications. Boca Raton:
Chapman and Hall/CRC.
5. Erdélyi, A. (Ed.). (1955). Higher transcendental functions (Vols. I–III). New York: McGraw-
Hill Book Company
6. Jumarie, G. (2009). Laplaces transform of fractional order via the Mittag-Leffler function and
modified Riemann-Liouville derivative. Applied Mathematics Letters, 22, 1659–1664.
7. Post, E. L. (1930). Generalized differentiation. Transactions of the American Mathematical
Society, 32, 723–781.
www.pdfgrip.com
8. Poularikas, A. D. (Ed.). (2000). The transforms and applications handbook (2nd ed.). CRC
Press: Boca Raton.
9. Schiff, J. L. (Ed.). (1999). The laplace transform, theory and applications. New York: Springer.
10. Wheeler, N. (1997). Construction and physical application of fractional calculus. Technical
report, Reed College Physics Department.
www.pdfgrip.com
Chapter 4
Fractional Differential Equations
4.1 The Existence and Uniqueness Theorem for Initial Value

Problems
Definition 1 Let be the fractional differential equation (FDE)
(Daα+ y)(t) = f [t, y(t)], α > 0, t > a,
with the conditions:

α−k
(Da+ y)(a+) = bk , k = 1, . . . , n,
called also Riemann–Liouville FDE.

Definition 2 Let the FDE
α
(Da+ y)(t) = f [t, y(t)], α > 0, t > a,
with the initial conditions:
(D k y)(0) = bk , k = 0, 1, . . . , n − 1,
called also Caputo FDE.

Lemma 1 Let y(t) be a function with continuous derivative in the interval Ih (0) =
[0, h] with values in [y0 − η, y0 + η], then y(t) satisfies the Caputo type
D α y(t) = f (t, y(t)), 0 < α ≤ 1, t > 0,
y(0) = y0 ,

www.pdfgrip.com
48 4 Fractional Differential Equations
if and only if it satisfies the Voltera1 integral,

t
1
y(t) = y0 + (t − u)α−1 f (u, y(u)) du.
(α) 0
Proof Let L[y(t] = Y be the LT of y(t). We have
s α Y − s α−1 y0 = L[f (t, y(t))],
y0 1
Y =
+ α L[f (t, y(t))],
s s
t
1
y(t) = y0 + (t − u)α−1 f (u, y(u)) du.
(α) 0
Definition 3 (Chebyshev2 Norm) The Chebyshev norm on a set S is:
f ∞ = sup{|f (x)| : x ∈ S},
where Supremum (sup) denotes the supremum.
Lemma 2 (The Weierstrass Test) Suppose that {fn (t)} is a sequence of real
functions defined on a set A, and there is a sequence of positive numbers {Rn }
satisfying:
∞

∀n > 1, ∀t ∈ A, |fn (t)| ≤ Rn , Rn < ∞.
n=1
∞

Then the series fn (t) is convergent.
n=1
Theorem 1 (Existence and Uniqueness for the Caputo Problem) Let a Caputo
FDE be
D α y(t) = f (t, y(t)), 0 < α ≤ 1, t > 0,
with the initial condition:
y(0) = y0 .
We consider the domain
D = [0, η] × [y0 − η, y0 + η],
1 V. Voltera (1860–1940).
2 P.L. Chebyshev (1821–1894).
www.pdfgrip.com
4.1 The Existence and Uniqueness Theorem for Initial Value Problems 49
on which f satisfies:
– f (t, y) is continuous,
– |f (t, y)| < M, where M = max |f (t, y)|, and Maximum (max) denotes the
(t,y)∈D
maximum function
– f (t, y) satisfy in D the Lipschitz3 condition in y if there is a constant K such
that:
|f (t, y2 ) − f (t, y1 )| ≤ K|y2 − y1 |.
Then it exists δ > 0 and a function y(t) ∈ C[0, η] unique for
1/α
η(α + 1)
δ = min η, ,
M
where Minimum (min) denotes the minimum function.

Proof We consider the Voltera integral (see Lemma 1)
t
1
y(t) = y0 + (t − u)α−1 f (u, y(u)) du,
(α) 0
and successive approximations:

t
1
yn (t) = y0 + (t − u)α−1 f (u, yn−1 (u)) du.
(α) 0
Using the method of successive approximations, on the basis of the Weierstrass

test we prove the existence and the uniqueness of the solution of Caputo FDE.
For the sequence {yn (t)}, we can prove that:
(i) the sequence {yn (t)} is well defined,
(ii) the sequence is uniformly continuous,
(iii) and its limit y(t) is unique.
Proof
(i) We will use the induction method. In the case n = 0 it is obvious.
If n = 1, then we have:
1 t M t

|y1 (t) − y0 | = (t − u)α−i f (u, y0 ) du < (t − u)α−1 du
(α) 0 (α) 0
Mt α Mδ α

= ≤ < η.
(α + 1) (α + 1)
3 R.O.S. Lipschitz (1832–1903).

www.pdfgrip.com
If we assume
|yn−1 − y0 | ≤ η,
then it follows that:

1 t M t

|yn − y0 | = (t − u) f (u, yn−1 )du <
α−1
(t − u)α−1 du
(α) 0 (α) 0
Mt α Mδ α

= ≤ < η.
(α + 1) (α + 1)
(ii) We consider the series

∞

y0 + (yk+1 (t) − y(t)),
k=0
equal with:

n−1
y0 + (yk+1 (t) − yk (t)) = yn+1 (t).
k=0
We have:
1 t

|y2 − y1 | = (t − u)α−1 [f (u, y1 (u)) − f (u, y0 )] du
(α) 0
K t KM

≤ (t − u)α−1 |y1 − y0 |du ≤ tα
(α) 0 (α + 1)
KM

≤ δ α = Kη.
(α + 1)
1 t

|y3 − y2 | = (t − u)α−1 [f (u, y2 (u)) − f (u, y1 )]du
(α) 0
K t K 2η

≤ (t − u)α−1 |y2 − y1 |du ≤ tα
(α) 0 (α + 1)
K 2η K 2 η2

≤ δα = ,
(α + 1) M
K n ηn
|yn+1 − yn | ≤ .
M n−1
www.pdfgrip.com
Using the Weierstrass test [9, 10] we obtain:

∞
∞
1 Kη n
| . . . | = y0 + η + .
M M
n=0 n=1
M
The series are convergent for η < .
K
Thus the sequence {yn (t)} is uniform convergent on the compact [0, η].
Hence, yn (t) is convergent to a function y(t) for t ∈ [0, η].
∀η > 0, ∃N positive number so for n > N we have:
|yn (t) − y(t)| < η.
This limit is unique.

(iii) Let x(t) be another limit for {yn (t)}, then:
|x(t) − y(t)| = |x(t) − yn (t) + yn (t) − y(t)|

η η
≤ |yn (t) − x(t)| + |yn (t) − y(t)| ≤ + = η.
2 2
Remark (Another Solution) In order to prove the existence of the solution we can
introduce the set
U = {y ∈ C[0, η] : y − y0 ≤ η}
and an operator A:
t
1
Ay(t) = y0 + (t − u)α−1 f (u, y(u)) du,
(α) 0
where A has a fixed point, and U is a closed and convex subset of all continuous
functions on [0, η] equipped with Chebyshev norm [12].
Generally:

n t
bj 1 f (u, y(u))
y(t) = (t − a)α−j + du,
(α − j + 1) (α) 0 (t − u)1−α
j =0
where t > 0, n − 1 ≤ α < n.

The technique used for proving the existence solution of the Voltera equation is
often the successive approximation:

n
bk
y0 (t) = t α−k ,
(α − k + 1)
k=1
www.pdfgrip.com
t
1
yi (t) = y0 (t) + (t − τ )α−1 f (τ, yi−1 (τ ))dτ, i = 1, 2, . . .
(α) 0
y(t) = lim yi (t).

i→∞
Example 1 Using method of successive approximation we solve the FDE:
D α y(t) = t 2 + y 2 , 0 < α ≤ 1, y(0) = 0, (t, y) ∈ [−1, 1] × [−1, 1].
Solution
– For α = 1, we have method of successive approximation or the method of
Picard.4
We construct a sequence {yn (t)} by the recurrence
t
yn (t) = y0 + f [u, yn−1 (u)]du, n = 1, 2, . . .
0
The {yn (t)} is convergent to an exact solution of the equation
y (t) = f [t, y(t)] = t 2 + y 2 , y(0) = 0,
in some interval 0 − h < t < 0 + h in the rectangle |t − t0 | ≤ a = 1, |y − y0 | ≤

b = 1,

b
h = min a, , M = max |f (t, y)|,
M (t,y)∈D
yn (t) is given by the inequality
MN n−1 n ∂f

|y(t) − yn (t)| ≤ h , N = max .
n! (t,y)∈D ∂y
For
1
M = 2, a = 1, h = ,
2
it results:
y0 (t) = 0,
t
t3
y1 (t) = (u2 + y02 )du = ,
0 3
4 E. Picard (1856–1941).
www.pdfgrip.com
t t3 t7
y2 (t) = + ,
(u2 + y12 )du =
3 63
0 t
t3 t7 2t 1 1 t 15
y3 (t) = (u + y1 )du =
2 2
+ + + ,
0 3 63 2079 59535
3
2 1 1
|y3 (t) − y(t)| ≤ 22 = , N = max |2y| = 2.
3! 2 6
– For 0 < α ≤ 1, we obtain:
y0 = 0,
t
1
yn (t) = (t − u)α−1 [u2 + yn−1
2
(u)]du.
(α) 0
We can calculate y1 (t):

t
1
y1 (t) = (t − u)α−1 u2 du.
(α) 0
The LT of this convolution is

1 1 (α) (3) (3)
Y1 = L[uα−1 ]L[u2 ] = = α+3 ,
(α) (α) s α s 3 s
from which, by inversion, we obtain:
2t α+2
y1 (t) = .
(α + 3)
Similarly, we obtain also:

t
1 u2α+4
y2 (t) = (t − u)α−1 u2 + 4 du,
(α) 0 2 (α + 3)
with:
2 4 (2α + 5)
Y2 (s) = + ,
s α+3 2 (α + 3) s 3α+5
and finally:
2t α+2 4(2α + 5) t 3α+4

y2 (t) = + 2 ,
(α + 3) (α + 3) (3α + 5)
www.pdfgrip.com
t3 t7
+ .
For α = 1, we obtain y2 (t) =
3 63
Example 2 Using method of successive approximation we solve the FDE:
D α y(t) = 1 + ty(t) + y 2 (t), 0 < α ≤ 1, y(0) = 0.
Solution As in the previous example, we have:

y0 (t) = 0, t
1
yn (t) = (t − u)α−1 1 + uyn−1 (u) + yn−1
2
(u) du,
(α) 0
1 t
y1 (t) = (t − u)α−1 du,
(α) 0
1 1 (α) 1 (1)
Y1 = L[uα−1 ]L[1] = α
= α+1 ,
(α) (α) s s s
tα
y1 (t) =
(α + 1)
t
1
y2 (t) = (t − u)α−1 1 + u · y1 (u) + y12 (u) du
(α) 0
1 (2α + 1) 1 α+1
Y2 = (α + 1) + + 2α+2 ,
s α+1 (α + 1) s
2 3α+1 s
tα (2α + 1) t 3α t 2α+1
y2 = + 2 + (α + 1) .
(α + 1) (α + 1) (3α + 1) (2α + 1)
4.2 Linear Fractional Differential Equations
A linear FDE is an equation of form
(D αn + an−1 D αn−1 + . . . a1 D α1 + a0 )y(t) = f (t), α ∈ R,
y (k) (0) = bk , k = 0, 1, 2, . . ., n − 1.
An equation which is not linear is called nonlinear.

Theorem 2 (Existence and Uniqueness) If f (t) is bounded on (0, T ) and ak =
ak (t), k ∈ {0, 1, . . ., n − 1} are continuous functions on [0, T ], the equation has a
unique solution.
www.pdfgrip.com
4.2 Linear Fractional Differential Equations 55
Proof The proof used here will be based on the proof of the existence and
uniqueness of the solution of Voltera integral equation.
Theorem 3 The linear FDE:
D α y(t) = f (t), where: n − 1 < α < n,
and
y (k) (0) = 0, k = 0, 1, 2, . . . , n − 1,
has the solution:

t
1
y(t) = (t − u)α−1 f (u) du.
(α) 0
Proof We apply the LT:
L[y(t)] = Y = Y (s),

L D α y(t) = L[f (t)] = F (s),
F (s)
s α Y = F (s) ⇒ Y =
sα
and using the convolution theorem it results the assumption of this theorem.
D α y(t) = λy(t), where: n − 1 < α < n,
with the initial condition
y (k) (0) = bk , bk ∈ R, k = 0, 1, 2, . . . , n − 1,
has the solution:

n−1
y(t) = bk t k Eα,k+1 (λt α ).
k=0
Proof We apply the LT method:
L[y(t)] = Y = Y (s),

L D α y(t) = λL[y(t)],
www.pdfgrip.com

n−1
sα Y − s α−k−1 y (k) (0) − λY = 0,
k=0
n−1

n−1 α−k−1
s
n−1
Y = bk = L bk t Eα,k+1 (λt ) = L
k α k α
bk t Eα,k+1 (λt ) .
sα − λ
k=0 k=0 k=0
Because L[y(t)] = Y it results the statements of the theorem.

D α y(t) = f (t), where: 0 < α < 1,
with the initial condition
y(0) = A,
and where:
∞
f (n) (0)
f (t) = t n,
n!
n=0
has the solution:

∞
f (n) (0)
y(t) = A + t α t n.
(n + α + 1)
n=0
Proof We apply the LT:
L[y(t)] = Y = Y (s),

L D α y(t) = s α Y − As α−1 ,
∞
f (n) (0)
s α Y − As α−1 = L[f (t)] = L[t n ],
n!
n=0
∞
∞

y(t) = yn , Y = L[y(t)] = Yn ,
n=0 n=0
∞
∞
A 1 f (n) (0)
Yn = + α L[t n ],
s s n!
n=0 n=0
www.pdfgrip.com
A
Y0 = ⇒ y0 = A,
s
f (n) (0) (n + 1) f (n) (0)

Yn = ⇒ yn = ,
(n + 1) s n+α+1 (n + α + 1)
∞
f (n) (0)
y(t) = A + t n+α .
(n + α + 1)
n=0
aD α y(t) + by(t) = f (t), 0 < α < 1,
y(0) = A,
where:
∞
f (n) (0)
f (t) = t n,
n!
n=0
has the solution:

∞
b α f (n) (0) n b α
y(t) = AEα,1 − t + t 1 − Eα,n+1 (− t ) .
a a a
n=0
Proof Applying the LT, it results:
L[y(t)] = Y = Y (s),

L D α y(t) = s α Y − As α−1 ,
∞
f (n) (0)
s α Y − As α−1 + bY = L[f (t)] = L[t n ],
n!
n=0
∞
∞

y(t) = yn , Y = L[y(t)] = Yn ,
n=0 n=0
∞
∞

As α−1 1 f (n) (0)
Yn = α + α L[t n ],
as + b as + b n!
n=0 n=0
www.pdfgrip.com
As α−1 A 1
Y0 = ⇒ Y0 = ,
as α + b as b
1+
as α
1 f (n) (0) f (n) (0) 1

Yn = L[t n ] ⇒ Yn = n+1+α ,
as α+ b n! as b
1+
as α

1 b 1 b2 1
Y0 = A − + + . . . ,
s a s α+1 a 2 s 2α+1

b tα b2 t 2α
y0 = A 1 − + + ... ,
a (α + 1) a 2 (2α + 1)

b
y0 = AEα,1 − t α ,
a

f (n) (0) 1 b 1 b2 1
Yn = − + 2 n+1+3α + . . . ,
a s n+1+α a s n+1+2α a s

f (n) (0) t n+α b t n+2α b2 t n+3α
yn = − + 2 + ... ,
a (n + 1 + α) a (n + 1 + 2α) a (n + 1 + 3α)

f (n) (0) n b α
yn = t 1 − Eα,n+1 − t .
a a
Finally, we obtain the solution:

∞

y(t) = y0 + yn ,
n=0
∞
b f (n) (0) n b
y(t) = AEα,1 − t α + t 1 − Eα,n+1 − t α .
a a a
n=0
Example 1 We will establish here the solution of the FDE:
D α y(t) + y(t) = 1,
y(0) = 0.
www.pdfgrip.com
Solution We apply the LT:

L D α y(t) + L[y(t)] = L[1],
1
s α Y − s α−1 y(0) + Y = ,
s
1
Y = ,
s(s α+ 1)
1 1
Y = ,
s α+1 1
1+
sα
and using the identity
1
= 1 − u + u2 − u3 + . . . |u| < 1,
1+u
we obtain:
1 1 1
Y = − 2α+1 + 3α+1 + . . .
s α+1 s s
Finally, it results:
tα t 2α t 3α
y(t) = − + + ...
(α + 1) (2α + 1) (3α + 1)
For α = 1, we can use the Maple or Mathematica commands in order to establish

the solution:
MAPLE
ec: = diff(y(t),t) + y(t) = 1;

dsolve({ec,y(0) = 0},y(t),type = series);
MATHEMATICA
Clear["‘*"]c
ec := y’[t] + y[t] == 1;
sol = DSolve[{ec, y[0] == 0}, y, t]
Series[y[t] /. sol, {t, 0, 10}]
It results the solution:
t2 t3 t4
y(t) = t − + − + ...
3 6 24
www.pdfgrip.com
Example 2 We consider the FDE
D α y(t) = y(t),
y(0) = 0, y (0) = 1.
We will establish the solution of this equation for the cases:

1. 1 < α ≤ 2,
2. 2 < α ≤ 3.
Solution
1. For the case 1 < α ≤ 2, using the LT, we have:
L[D α y(t]) = L[y(t)],
s α Y − s α−1 y(0) − s α−2 y (0) = Y,
s α−2
Y = ,
sα − 1
1 1
Y = ,
s2 1
1−
sα
and using the identity
1
= 1 + u + u2 + u3 + |u| < 1,
1−u
we have
1 1 1
Y = + α+2 + 2α+2 + . . .,
s2 s s
and the solution:
t t α+1 t 2α+1
y(t) = + + + ...
(2) (α + 2) (2α + 2)
We can note that:
t t3 t5
lim y(t) = + + . . . = sinh(t).
α→2 1! 3! 5!
www.pdfgrip.com
2. For the case 2 < α ≤ 3, using same procedure, we have:
s α Y (s) − s α−1 y(0) − s α−2 y (0) − s α−3 y (0) = Y (s).
For y (0) = b, we obtain:
s α−2 s α−3
Y = + b ,
sα − 1 sα − 1
and using the residues theorem we have:
s+b
lim Y = ,
α2 s(s 2 − 1)
r1 = Res(Y est ) = lim sY est = −b,

0 s→0
(b − 1)e−t
r2 = Res(Y est ) = lim (s + 1)Y est = ,
−1 s→−1 2
(b + 1)et
r3 = Res(Y est ) = lim (s − 1)Y est = .
1 s→1 2
The solution will be:
y(t) = r1 + r2 + r3 = sinh(t) + b cosh(t) − b.
Observation
This equation can be solved also in terms of perturbation method. In this case we
take
y(t) = t + c1 t α+1 + c2 t 2α+1 + c3 t 3α+1 + . . . ,
and using the formula:

⎧
⎪ (β + 1) β−α
⎪
⎨ t β >n−1
(β + 1 − α)
D∗α t β =
⎪
⎪
⎩
0 β ≤ n − 1.
D∗α y(t) = c1 D∗α t α+1 + c2 D∗α t 2α+1 + c3 D∗α t α+1 + . . .

= t + c1 t α+1 + c2 t 2α+1 + c3 t 3α+1 + . . .
www.pdfgrip.com
we have
(α + 2) (2α + 2) α+1 (3α + 2) 2α+1
c1 t + c2 t + c3 t + ...
(2) (α + 2) (2α + 2)
= t + c1 t α+1 + c2 t 2α+1 + c3 t 3α+1 + . . .
and after the identification we obtain:

1
c1 (α + 2) = 1 ⇒ c1 = ,
(α + 2)
(2α + 2) 1
c2 = c1 ⇒ c2 = ,
(α + 2) (2α + 2)
(3α + 2) 1
c3 = c2 ⇒ c3 = ,
(2α + 2) (3α + 2)
...
For α = 2, we have:
y (t) = y(t).
We apply the LT:
L[y (t]) = L[y(t)], L[y(t)] = Y = Y (s),
s 2 Y − sy (0) − y(0) = Y,
1
Y = ,
s2 − 1
est e−t
r1 = Res Y est = lim Y est = lim (s + 1) =− ,
−1 s→−1 s→−1 (s − 1)(s + 1) 2
est et
r2 = Res Y est = lim Y est = lim (s − 1) = ,
1 s→1 s→1 (s − 1)(s + 1) 2
et − e−t
f (t) = r1 + r2 = = sinh(t).
2
We can use here the MAPLE commands:
with(inttrans);
ec:=diff(y(t),t$2) = y(t);
dsolve(ec,D(y)(0) = 1,y(0) = 0,y(t), method = laplace);
www.pdfgrip.com
Example 3 Find the solution of the FDE:
D 2 y(t) − D 3/2 y(t) − y(t) + t + 1 = 0,
y(0) = y (0) = 1.
Solution We apply the LT:

3
L D 2 y(t) − L D 2 y(t) − L[y(t)] + L[t] + L[1] = 0,

L D 2 y(t) = s 2 Y − sy(0) − y (0) = s 2 Y − s − 1,
3 3 1 1 1 1 1
L D 2 y(t) = s 2 Y − s 2 y(0) − s − 2 y (0) = s 2 Y − s 2 − s − 2 =
s2 − s − 1
= √ ,
s
L[y(t)] = Y,
1
L[t] = ,
s2
1
L[1] = ,
s
s2Y − s − 1 2
s2Y − s − 1 − √ − s Y − s − 1 = 0,
s
1 1
s 2 Y − s − 1 = 0, ⇒ Y = + 2,
s s
y(t) = L−1 [Y ], ⇒ y(t) = 1 + t.
Example 4 We consider the problem [11], with the initial condition:
d 1/2 y y
= , y(0) = 0.
dx 1/2 x
which can be rewritten as:
tD 1/2 y(t) − y(t) = 0, y(0) = 0.
We will establish the solution of this equation.

www.pdfgrip.com
Solution We apply the LT method:

L tD 1/2 y(t) − L[y(t)] = 0,
d 1/2
L tD 1/2 y(t) = − s Y − s −1/2 y(0) − Y = 0,
ds

dY 1 1
+ + √ Y = 0.
ds 2s s
We obtain:
√
e−2 s Ce−1/t
Y (s) = C √ ⇒ y(t) = √ .
s πt
The same solution can be found using the Maple program:

MAPLE
with(inttrans):
ec:= diff(Y(s),s) + (1/(2*s)+1/sqrt(s)) - Y(s) = 0;
F(s):= dsolve(ec,Y(s));
f(t):= invlaplace(F(s),s,t);
Other applications can be found in [5, 6].
4.3 Nonlinear Equations
4.3.1 The Adomian Decomposition Method
The Adomian5 method [1–3], applied to the ordinary and partial differential
equations of integer order was extended also to the case of FDE (for further details
and examples see [7, 8]).
Adomian Polynomials
We will denote these polynomials by A0 , A1 , . . . An , . . ..
We consider a nonlinear analytic function G(y(t), t) and that y(t0 ) = y0 , in
the D domain. The Adomian method consists in the decomposition the unknown
function y(t) in a series of form y(t) = y0 + y1 + y2 + . . . + yn + . . ., where yn can
be expressed in terms of Adomian polynomials An .
5 G. Adomian (1922–1996).
www.pdfgrip.com
4.3 Nonlinear Equations 65
The Adomian polynomials are defined [1, 2, 8]:

⎡ ⎤
1 dn ⎣
n
An = G(t, yj λj )⎦ .
n! dλn
j =0 λ=0
These polynomials can be established algorithmically, using the symbolic pro-

gramming packages, with the aid of do . . . end do repetition statements:
Step 1: To calculate first the Adomian polynomial A0 , i.e., A0 = G(y0 , t).
Step 2: Iterative calculation of Ak using the for . . . do .. end do loop:
> for k = 0 to n - 1 do
> Ak = Ak (y0 + λ ∗ y1 , . . . , yk + (k + 1) ∗ λ ∗ yk+1 )
> end do;
It must be underlined that in the Ak polynomial yi is replaced with
yi → yi + (i + 1) ∗ yi+1 ∗ λ, for: i = 0, 1, . . . , k.
Step 3:

d
Ak = (k + 1) ∗ Ak .
dλ λ=0
Step 4: We obtain, finally A0 , A1 , . . . , An and y0 , y1 , y2 , . . . in terms of An .

A given f (u) can be expressed as a series of An ,
∞

f (u) = An ,
n=0
∞

and u = un .
n=0
∞

The series An can be rearranged as a generalized Taylor series:
n=0
∞

f (u) = An = f (u0 ) + (u1 + u2 + . . .)f (1) (u0 )
n=0

u21
+ + u1 u2 + . . . f (2) (u0 ) + . . .
2!
∞
∞

(n)
= (u − u0 ) /n! f (u0 ) =
n
(u1 + u2 + . . .)n /n! f (n) (u0 ),
n=0 n=0
www.pdfgrip.com
so that:
A0 = f (u0 ),
A1 = u1 f (1) (u0 ),
A2 = u2 f (1) (u0 ) + (1/2!)u21 f (2) (u0 ),
A3 = u3 f (1) (u0 ) + u1 u2 f (2) (u0 ) + (1/3!)u31 f (3) (u0 ),
...
Example 1 We will establish the Adomian polynomials for G(y) = y 2 .

The Adomian polynomials will be:
A0 = y02 , A1 = 2y0 y1 , A2 = y12 + 2y0 y2 , A3 = 2y1 y2 + 2y0 y3 , etc.
For calculation you can use also the following Maple, or Mathematica sequences:
MAPLE
restart;
with(LinearAlgebra):
unassign(’y,lambda’):
f:=y->y^2:
S:=lambda->sum(y[i]*lambdaî,i=0..4):
g:=lambda->(S(lambda))^2:
c:=Vector(4,n->diff(1/n!*g(lambda),lambda$n):
A:=<subs(lambda=0,g(lambda)),subs(lambda=0,c)>;
MATHEMATICA
Clear["‘*"]
f[y_] := y^2;
S[\[Lambda]_] := Sum[y[i]*\[Lambda]î, {i, 0, 5}];
g[\[Lambda]_] := f[S[\[Lambda]]];
ad = Table[
1/n!*D[g[\[Lambda]], {\[Lambda],n}]/. \[Lambda]-> 0,
{n, 0, 5}] // Simplify;
TableForm[ad, TableAlignments -> Left]
Example 2 Let us calculate the Adomian polynomials for G = y 3 .
Solution
A0 = y03 , A1 = 3y1 y02 , A2 = 3y02 y2 + 3y12 y0 ,
A3 = y13 + 6y0 y1 y2 + 3y02 y1 , etc.
The basic principles of this algorithm remain unchanged for other definitions of the
function G.
www.pdfgrip.com
Example 3 Let us calculate the Adomian polynomials for G = f (u).

We obtain successively:
A0 = f (u0 ), A1 = u1 (d/du0 )f (u0 ),
A2 = u2 (d/du0 )f (u0 ) + (u21 )/2!)(d 2 /du20 )f (u0 ),
A3 = u3 (d/du0 )f (u0 ) + (u1 u2 )(d 2 /du20 )f (u0 ) + (u31 /3!)(d 3 /du30 )f (u0 ), etc.
Example 4 Find the Adomian polynomials for G = sin θ . It results:
A0 = sin θ0 , A1 = θ1 cos θ0 , A2 = −(θ12 /2) sin θ0 + θ2 cos θ0 , etc.
4.3.2 Decomposition of Nonlinear Equations
We consider the nonlinear FDE of type:
D α y(t) + Ry(t) + Ny(t) = f (t), y (k) (0) = ck , k = 0, 1, . . . , n − 1, α > 0,
where N is a nonlinear operator, and Ry is a residual part of the equation.

We apply the LT to the equation. It follows:
L[D α y(t)] = s α Y − s α−1 y(0) − s α−2 y (0) − . . . − y (n−1) (0) = s α Y − c,
c = s α−1 y(0) + s α−2 y (0) − . . . + y (n−1) (0),
where c is a constant.
We use the following decomposition of y(t)
∞

y(t) = yn (t),
n=0
with
∞

Ny(t) = An ,
n=0
www.pdfgrip.com
where An are Adomian polynomials:

∞
1 dn
An = N i
λ yi ,
n! dλn
i=0
∞
∞ ∞
c 1 1
L yn = αY − αL R yn − α L An .
s s s
n=0 n=0 n=0
We have after calculations:

c 1
Y0 = L[y0 ] = α
+ α L [f (t)] ,
s s
1 1
Y1 = L[y1 ] = − L[Ry0 ] − α L[A0 ],
sα s
1 1
Y2 = L[y2 ] = − α
L[Ry1 ] − α L[A1 ],
s s
...
1 1
Yn = L[yn ] = − L[Ryn−1 ] − α L[An−1 ].
sα s
Example 1 Solve the nonlinear FDE using the Adomian decomposition method:
D α y(t) = t + y 2 , 1 < α ≤ 2,
y(0) = 0, y (0) = 1.
Solution In order to solve the equation we apply the LT:

L[D α y(t)] = L t + y 2 ,
L[y(t)] = Y,
L[D α y(t)] = s α Y − s α−1 y(0) − s α−2 y (0) = s α Y − s α−2 ,
1 1
Y = + L t + y 2
,
s2 sα
so that, for the decomposition
∞

y(t) = yn (t),
n=0
www.pdfgrip.com
we obtain:
∞
∞

Y = Yn , t + y2 = An ,
n=0 n=0
where An are Adomian polynomials. We obtain:
A0 = t + y02 , A1 = 2y0 y1 , A2 = y12 + 2y0 y2 , A3 = 2y1 y2 + 2y0 y2 , . . .
∞
∞
1 1
Yn = + An ,
s2 sα
n=0 n=0
1
Y0 = ⇒ y0 = t,
s2
A0 = t + t 2 ,
1 1
Y1 = L[A0 ] ⇒ Y1 = L t + t2 ,
sα s α
1 1 t α+1 t α+2
Y1 = +2 ⇒ y1 = +2 ,
s α+2 s α+3 (α + 2) (α + 3)
A1 = 2y0 y1 ,
t α+2 t α+3
A1 = 2 +4 ,
(α + 2) (α + 3)
1 2(α + 2) 4(α + 3)
Y2 = α
L[A1 ] ⇒ Y2 = 2α+3 + 2α+4 ,
s s s
2(α + 2)t 2α+2 4(α + 3)t 2α+3

y2 = + ,
(2α + 3) (2α + 4)
y(t) = y0 (t) + y1 (t) + y2 (t) + . . . ,
and finally, it yields:
t α+1 t α+2 2(α + 2)t 2α+2 4(α + 3)t 2α+3

y(t) = t + +2 + + + ...
(α + 2) (α + 3) (2α + 3) (2α + 4)
For α = 2, we obtain:
y (t) = t 2 + y 2 (t), y(0) = 0, y (0) = 1.

www.pdfgrip.com
The result
t3 t4
y(t) = t + + + ...,
6 12
can be obtained also on computer, using the sequences:
MAPLE
ec: = diff(y(t), t$2) = t + (y(t))^2;

dsolve({ec,y(0) = 0,D(y)(0) = 1},y(t),type = series);
MATHEMATICA
ec:=y"[t] == t + y[t]*y[t];
sol=DSolve[{ec,y[0]==0,y’[0]==1},y,t];
Series[y[t]/.sol,{t,0,10}]
MATHEMATICA
Clear["‘*"]
Manipulate[
f[t_] := t + t^3/6 + t^4/12 + t^5/120;
y[t_] :=
t + t^(a + 1)/Gamma[a + 2] + 2*t^(a + 2)/Gamma
[a + 3] + 2*(a + 2)*t^(2*a + 2)/Gamma[2*a + 3] +
4*(a + 3)*t^(2*a + 3)/Gamma[2*a + 4];
Plot[{f[t], y[t]}, {t, 0, 1}, ImageSize -> 300,
Frame -> True], {{a, 1/2}, 0, 1}]
The functions f (t) and y(t) calculated here are plotted in Fig. 4.1.
Example 2 Let us solve the FDE:
D α y(t) = 1 + y 2 (t), 0 < α ≤ 1,
where:
y(0) = 0,
using the Adomian decomposition method.
Solution To solve this problem we apply the LT:

L[D α y(t)] = L[1] + L y 2 ,
L[y(t)] = Y,
www.pdfgrip.com
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Fig. 4.1 Plots of the functions f (t) and y(t) from the Example 1
L[D α y(t)] = s α Y − s α−1 y(0) = s α Y,
1 1 2
Y = + L y .
s α+1 sα
For the decomposition
∞

y(t) = yn (t),
n=0
we obtain
∞
∞

Y = Yn , y2 = An ,
n=0 n=0
where An are the Adomian polynomials:
A0 = y02 , A1 = 2y0 y1 , A2 = y12 + 2y0 y2 , A3 = 2y1 y2 + 2y0 y2 , . . .

www.pdfgrip.com
∞
∞
1 1
Yn = + An ,
s α+1 sα
n=0 n=0
1 tα
Y0 = ⇒ y0 = ,
s α+1 (α + 1)
t 2α
A0 = y02 = ,
2 (α + 1)

1 1 t 2α
Y1 = L[A0 ] ⇒ Y1 = L ,
sα sα 2 (α + 1)
1 (2α + 1)
Y1 = ⇒ y1 = L−1 Y1 ,
s 3α+1 2 (α + 1)
(2α + 1) t 3α
y1 = ,
2 (α + 1) (3α + 1)
A1 = 2y0 y1 ,
1
Y2 = L[A1 ],
sα
(2α + 1) (4α + 1) 1
Y2 = 2 ,
3 (α + 1) (3α + 1) s 5α+1
y2 = L−1 [Y2 ],
(2α + 1) (4α + 1)
y2 = 2 t 5α ,
3 (α + 1) (3α + 1)(5α + 1)
y(t) = y0 (t) + y1 (t) + y2 (t) + . . .
tα (2α + 1) t 3α
y(t) = + 2
(α + 1) (α + 1) (3α + 1)
(2α + 1) (4α + 1)
+2 t 5α + . . .
3 (α + 1) (3α + 1)(5α + 1)
You can also use the programs:

MAPLE
ec:=diff(y(t),t) = 1 + y(t))^2;
dsolve({ec,y(0) = 0},y(t),type = series);
www.pdfgrip.com
MATHEMATICA
ec:=y’[t] == 1 + y[t]*y[t];
sol=DSolve[{ec,y[0]==0},y,t];
Series[y[t]/.sol,{t,0,10}]
Finally, we have:
t3 2t 5
y(t) = t + + + ...
3 15
Example 3 Solve the Ghelfand’s6 FDE:
D 2α y(t) = 2ey(t) , 0 < α ≤ 1,
where:
y(0) = y α (0) = 0,
using the Adomian decomposition method.

Solution To solve this problem we apply the LT:

L[D 2α y(t)] = 2L ey(t) ,
y α (0) α
L[y(t)] = Y, y0 (t) = y(0) + t = 0,
(α + 1)
L[D 2α y(t)] = s 2α Y − s 2α−1 y(0) = s α Y,

∞

L e y(t)
=L An .
n=0
For the decomposition

∞

y(t) = yn (t),
n=0
we obtain
∞
∞

Y = Yn , y2 = An ,
n=0 n=0
6 I.M. Ghelfand (1913–2009).

www.pdfgrip.com
where An are the Adomian polynomials:
A0 = ey0 , A1 = y1 ey0 , . . .
∞
∞

2
Yn = L An ,
s 2α
n=0 n=0
y0 = 0,
A0 = ey0 = 1,
1 2
Y1 = L[A0 ] ⇒ Y1 = ,
s 2α s 2α+1
2t 2α
y1 = ,
(2α + 1)
A1 = y1 ey0 ,
4
Y2 = ,
s 4α+1
y2 = L−1 [Y2 ],
4t 4α
y2 = ,
(4α + 1)
y(t) = y0 (t) + y1 (t) + y2 (t) + . . .
2t 2α 4t 4α
y(t) = 0 + + + ...
(2α + 1) (4α + 1)
It can be used also the program:

MATHEMATICA
Clear["‘*"]
Manipulate[
f[t_] := t^2 + t^4/6;
y[t_] := 2*t^(2*a)/Gamma[2*a + 1] + $*t^(4*a)/Gamma
[4*a + 1]; Plot[{f[t], y[t]}, {t, 0, 1},
ImageSize -> 300, Frame -> True], {{a, 1/2}, 0, 1}]
Figure 4.2 shows the plots of the functions f (t) and y(t).
www.pdfgrip.com
0
0.0 0.2 0.4 0.6 0.8 1.0
Fig. 4.2 Plots of the functions f (t) and y(t) from the Example 3
4.3.3 Perturbation Method
Here, we will extend the perturbation method for the case of FDE with the aid of
some examples.
D α y(t) = 1 + y 2 (t), 0 < α ≤ 1,
for the initial condition:
y(0) = 0,
using the small parameter (perturbation) method, 0 < 1.

Solution We consider a solution of form:
y(t) = y0 (t) + y1 (t) + 2 y2 (t) + 3 y3 (t) + . . . ,

www.pdfgrip.com
which, replaced in the equation, gives:
D α y0 (t) = 1, y0 (0) = 0,
D α y1 (t) = y02 , y1 (0) = 0,
D α y2 (t) = 2y0 y1 , y2 (0) = 0,
...
We apply the LT:
L[D α y0 (t)] = L[1], y0 (0) = 0,
1 1 tα
s α Y0 − y0 (0) = , Y0 = ⇒ y0 (t) = ,
s s α+1 (α + 1)
L[D α y1 )] = L[y02 ], y1 (0) = 0,
1 (2α + 1) (2α + 1) 1
s α Y1 = , Y1 = ,
2 (α+ 1) s 2α+1 2 (α + 1) s 3α+1
and by inverse LT, we obtain:
(2α + 1) t 3α
y1 (t) = .
2 (α + 1) (3α + 1)
It results also:
L[D α y2 ] = L[2y0 y1 ], y2 (0) = 0,

tα (2α + 1) t 3α
s Y2 = 2
α
L ,
(α + 1) 2 (α + 1) (3α + 1)
(2α + 1)
s α Y2 = 2 L t 4α ,
3 (α + 1)(3α + 1)
(2α + 1) (4α + 1)
Y2 = 2 ,
3 (α + 1)(3α + 1) s 5α+1
(2α + 1) (4α + 1) 5α
y2 = 2 t ,
3 (α + 1)(3α + 1) (5α + 1)
The solution for this example:
y(t) = y0 (t) + y1 (t) + y2 (t) + . . . ,

www.pdfgrip.com
will be:
tα (2α + 1) t 3α
y(t) = + 2 +
(α + 1) (α + 1) (3α + 1)
(2α + 1) (4α + 1)t 5α

+2 + ...
3 (α + 1)(3α + 1) (5α + 1)
The plot of f (t) and y(t) is done with the following program and presented in
Fig. 4.3.
MATHEMATICA
Clear["‘*"]
Manipulate[
f[t_] := t + t^3/3 + 2/15*t^5;
y[t_] :=
tâ/Gamma[a + 1] +
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Fig. 4.3 Plots of the functions f (t) and y(t) from the above Example 1
www.pdfgrip.com
Gamma[2*a + 1]/(Gamma[a + 1])^2*t^(3*a)/Gamma

[3*a + 1] + 2 Gamma[2*a + 1]/((Gamma[a + 1])^3*Gamma
[3*a + 1])* Gamma[4*a + 1]/Gamma[5*a + 1]*t^(5*a);
Plot[{f[t], y[t]}, {t, 0, 1}, ImageSize -> 300,
Frame -> True], {{a, 1/2}, 0, 1}]
−y(t)
D α y(t) = , 0 < α ≤ 1,
1+
for the initial condition:
y(0) = cos(), 0 < 1,
using the small parameter (perturbation) method.

Solution We consider a solution of form:
y(t) = y0 (t) + y1 (t) + 2 y2 (t) + 3 y3 (t) + . . . ,
1
= 1 − + 2 − 3 + . . .
1+
1 2 1
cos() = 1 − + 4 + . . . ,
2! 4!
which, replaced in the equation, gives:
D α (y0 + y1 + 2 y2 + . . .) = −(y0 + y1 + 2 y2 + . . .)(1 − + 2 + . . .)
D α y0 = −y0 , y0 (0) = 1,
D α y1 = y0 − y1 , y1 (0) = 0,
1
D α y2 = −y2 + y1 + y0 , y2 (0) = − ,
2
...
We apply the LT:
L[D α y0 (t)] = −L[y0 ], y0 (0) = 1,
s α−1
s α Y0 − y0 (0)s α−1 = −Y0 , Y0 =
sα + 1
www.pdfgrip.com
L[D α y1 ] = L[y0 ] − L[y1 ], y1 (0) = 0,
1
L[D α y2 ] = −L[y2 ] + L[y1 ] + L[y0 ], y2 (0) = − ,
2
sα
(s α + 1)Y1 = Y0 , Y1 = ,
(s α + 1)2
s α−1 s α−1 1 sα
Y2 = + − ,
(s α + 1)3 (s α + 1)2 2 sα + 1
tα t 2α t 3α
y0 (t) = 1 − + − + ...,
(α + 1) (2α + 1) (3α + 1)
1 1
Y1 = ,
s α+1 1 2
1+ α
s
1 1 1
Y1 = −2 +3 + ...,
s α+1 s 2α+1 s 4α+1
and by inverse LT, we obtain:
tα t 2α t 4α
y1 (t) = −2 +3 + ...
(α + 1) (2α + 1) (4α + 1)
But:

s α−1 1 1 1 1
= 2α+1 2 − 2 · 3 α + 4 · 3 2α + . . . .
1 3 s 2 s s
s 3α 1+ α
s
It results also:

1 t 2α t 3α
y2 (t) = 2 −2· 3 + ...
2 (2α + 1) (3α + 1)

tα t 2α t 4α
+ −2 +3 + ...
(α + 1) (2α + 1) (4α + 1)

1 tα t 2α t 3α
− 1− + − + ... .
2 (α + 1) (2α + 1) (3α + 1)
www.pdfgrip.com
4.4 Fractional Systems of Differential Equations
4.4.1 Linear Systems
Examples Solve the system of FDE:

⎧
⎨ D α x(t) = D β y(t) + 1, x(0) = 1, 0 < α ≤ 1
,
⎩
D β y(t) = 2D α x(t) − 1, y(0) = 1, 0 < β ≤ 1.
Solution We apply the LT method:
L[x(t)] = X, L[y(t)] = Y,
L[D α x(t)] = s α X − s α−1 x(0) = s α X − s α−1 ,
L[D β y(t)] = s β Y − s β−1 y(0) = s β Y − s β−1 .
We obtain the system

⎧
⎪ 1
⎪
⎪ X= ,
⎨ s
⎪
⎪
⎪
⎩Y = 1 − 1 ,
s s β+!
with the solution:
⎧
⎪
⎪ x(t) = 1,
⎨
⎪
⎪ tβ
⎩ y(t) = 1 − .
(β + 1)
4.4.2 Nonlinear Systems

(A) Method of Successive Approximations
For the system of FDE:

⎧ α
⎨ D x(t) = f (t, y(t)), x(0) = x0 ,
⎩
D α y(t) = g(t, x(t)), y(0) = y0 ,
www.pdfgrip.com
4.4 Fractional Systems of Differential Equations 81
we can use the following successive approximations [4]:

⎧ t
⎪
⎪ 1
⎪
⎪ xn (t) = x0 + f (u, yn−1 (u))(t − u)α−1 du,
⎨ (α) 0
⎪
⎪ t
⎪
⎪ 1
⎩ yn (t) = y0 + g(u, xn−1 (u))(t − u)α−1 du.
(α) 0
Example We apply the successive approximation method for the system of FDE
with initial conditions:
⎧ α
⎨ D x(t) = 3.5y(t)(1 − y(t)), x(0) = 0.2,
⎩
D α y(t) = 4x(t)(1 − x(t)), y(0) = 0.2.
We have:
⎧ t
⎪
⎪ 3.5
⎪ xn (t) = 0.2 +
⎪ yn−1 (u) − yn−1
2
(u) (t − u)α−1 du,
⎨ (α) 0
⎪
⎪ t
⎪
⎪ 4
⎩ yn (t) = 0.2 + xn−1 (u) − xn−1
2
(u) (t − u)α−1 du,
(α) 0
⎧ t
⎪
⎪ 3.5
⎪
⎪ x1 = 0.2 + 0.2 − (0.2) 2
(t − u)α−1 du,
⎨ (α) 0
⎪
⎪ t
⎪
⎪ 4
⎩ y1 = 0.2 + 0.2 − (0.2)2 (t − u)α−1 du.
(α) 0
Using the theorem regarding the product of convolution, we obtain the following
three iterations:
⎧ tα
⎪
⎪ x1 (t) = 0.2 + 0.56 ,
⎪
⎨ (α + 1)
⎪
⎪ tα
⎪
⎩ y1 (t) = 0.2 + 0.56 .
(α + 1)
⎧ t
⎪
⎪ 3.5
⎪
⎪ x2 (t) = 0.2 + y1 (u) − y21 (u) (t − u)α−1 du,
⎨ (α) 0
⎪
⎪ t
⎪
⎪ 4
⎩ y2 (t) = 0.2 + x1 (u) − x21 (u) (t − u)α−1 du,
(α) 0
www.pdfgrip.com
⎧
⎪
⎪ x2 (t) = 0.2
⎪
⎪ tα t 2α (2α + 1) t 3α
⎪
⎪
⎪
⎪ +0.56 + 1.344 − 0.808 2 ,
⎨ (α + 1) (2α + 1) (α + 1) (3α + 1)
⎪
⎪
⎪
⎪ y2 (t) = 0.2
⎪
⎪
⎪
⎪ tα t 2α (2α + 1) t 3α
⎩ +0.64 + 1.344 − 0.7077 2 ,
(α + 1) (2α + 1) (α + 1) (3α + 1)
⎧ t
⎪
⎪ 3.5
⎪
⎪ x3 (t) = 0.2 + y2 (u) − y 2
(u) (t − u)α−1 du
⎨ (α) 0 2
⎪
⎪ t
⎪
⎪ 4
⎩ y3 (t) = 0.2 + x2 (u) − x22 (u) (t − u)α−1 du
(α) 0
...
For α = 0.9 we can apply the Maple and Mathematica programs:

MAPLE
> restart;
> with(inttrans):
> Digits:=5:
> x:=array(0..10):
> y:=array(0..10):
> x[0]:=0.2:
> y[0]:=0.2:
> for k from 1 to 5 do
> x[k]:=evalf(0.2+3.5*invlaplace(1/s^0.9*laplace
(y[k-1]-(y[k-1])^2,
t,s),s,t));
> y[k]:=evalf(0.2+4*invlaplace(1/s^0.9*laplace
(x[k-1]-(y[k-1])^2,
t,s),s,t));
> od:
> print([x[k],y[k]]) od:
x y
0.2 0.2
0.2 + 0.58230 t^(9/10) 0.2 + 0.66548t^(9/10)
0.2 + 0.58230 t^(9/10) + 0.80171 t^(9/5) - 0.62304
t^(27/10),
0.2 + 0.66548 t^(9/10) + 0.72536 t^(9/5) - 0.71204
t^(27/10)
. . . . .
www.pdfgrip.com
4.4 Fractional Systems of Differential Equations 83
MATHEMATICA
Clear["‘*"]
Array[x, 10]
Array[y, 10]
For[{n = 0, x[0] = 0.2, y[0] = 0.2}, n < 4,
n++, {x[n + 1] =
0.2 + 3.5 InverseLaplaceTransform[
1/s^0.9 LaplaceTransform[y[n] - (y[n])^2, t, s],
s, t]// FullSimplify,
y[n + 1] =
0.2 + 4 InverseLaplaceTransform[
1/s^0.9 LaplaceTransform[(x[n] - (x[n])^2), t, s],
s, t]//
FullSimplify, Print["x=", x[n], " , ", "y= ", y[n]]}]
x = 0.2, y=0.2
x = 0.2+0.582262 t^0.9 , y= 0.2+0.665443 t^0.9
(B) Method of Laplace’s Transform
We will illustrate this method on the function:
F (y) = y − y 2 .
First, we will decompose F in terms of Adomian’s polynomials

∞

F = An ,
n=0
where A0 = y0 − y02 and
φ1 (λ) = (y0 + λy1 ) − (y0 + λy1 )2 ,
φ1 = y1 − 2y1 (y0 + λy1 ),

1
A1 = φ1 (0),
1!
from which we obtain: A1 = y1 − 2y0 y1 .
In the case of next step we have:
φ2 (λ) = (y1 + 2λy2 ) − 2(y0 + λy1 )(y1 + 2λy2 ),
φ2 = 2y2 − 2y1 (y1 + 2λy2 ) − 2y2 (y0 + λy1 ),

1
A2 = φ (0),
2! 2
or finally: A2 = y2 − y12 − y2 y0 .
www.pdfgrip.com
We have also:
φ3 (λ) = (y2 + 3λy3 ) − (y1 + 2λy2 )2 − (y2 + 3λy3 )(y0 + λy1 ),
1
A3 = φ (0),
3! 3
...
We consider now a system described by the equations [8]:

⎧
⎨ L[D α x(t)] = 3.5L[y(t)(1 − y(t))], x(0) = 0.2,
⎩
L[D α y(t)] = 4L[x(t)(1 − x(t))], y(0) = 0.2,
with initial conditions and we apply the LT to this system. We have:
L[x(t)] = X, L[y(t)] = Y,
L[D α x(t)] = s α − x(0)s α−1 ,
L[D α y(t)] = s α − y(0)s α−1 ,
We consider the solutions:

∞
∞

X= Xn , Y = Yn .
n=0 n=0
After calculations we have:

∞ ∞

L [x(t)(1 − x(t))] = L An , L [y(t)(1 − y(t))] = L Bn ,
n=0 n=0
where An and Bn are Adomian’s polynomials.
∞
∞
0.2 3.5
Xn = + αL Bn
s s
n=0 n=0
∞
∞
0.2 4
Yn = + αL An
s s
n=0 n=0
0.2
X0 = ⇒ x0 = 0.2
s
www.pdfgrip.com
References 85
0.2
Y0 = ⇒ x0 = 0.2
s
⎧ 3.5 0.56 tα
⎪
⎪ X = L[B ] = , ⇒ x (t) = 0.56
⎪
⎨ 1
sα
0
s α+1
1
(α + 1)
⎪
⎪ tα
⎪
⎩ Y1 = 4 L[A0 ] = 0.64 , ⇒ y1 (t) = 0.64
sα s α+1 (α + 1)
tα
B1 = y1 − 2y0 y1 = y1 (1 − 2y0 ) = 0.64 (1 − 2 · 0.2)
(α + 1)
tα tα
= 0.64 · 0.6 · = 0.384
(α + 1) (α + 1)
tα
A1 = x1 − 2x0 x1 = x1 (1 − 2x0 ) = 0.56 (1 − 2 · 0.2)
(α + 1)
tα tα
= 0.56 · 0.6 · = 0.336
(α + 1) (α + 1)
3.5 1.344 t 2α
X2 = L[B 1 ] = ⇒ x2 (t) = 1.344
sα s 2α (2α + 1)
4 1.344 t 2α
Y2 = L[A1 ] = 2α ⇒ y2 (t) = 1.344
s α s (2α + 1)
Finally, the solution is:

⎧
⎪
⎪ tα t 2α
⎪
⎪ x(t)=x0 (t)+x1 (t)+x2 (t)+ . . . = 0.2 + 0.56 + 1.344 + ...
⎨ (α + 1) (2α + 1)
⎪
⎪
⎪
⎪ tα t 2α
⎩ y(t)=y0 (t)+y1 (t)+y2 (t) + . . . = 0.2 + 0.64 + 1.344 + ...
(α + 1) (2α + 1)
References
1. Adomian, G. (1988). A review of the decomposition method in applied mathematics. Journal

of Mathematical Analysis and Applications, 135(2), 501–544.
2. Adomian, G. (1994). Solving frontier problems of physics: The decomposition method.
Fundamental theories of physics. Dordrecht: Springer.
3. Cherruault, Y. (1989). Convergence of Adomian’s method. Kybernetes, 18(2), 31–38.
www.pdfgrip.com
4. El’sgol’ts, L. E., & Norkin, S. B. (1973). Introduction to the theory of differential equations
with deviating arguments. Mathematics in science and engineering. New York: Academic
Press.
5. Kazem, S. (2013). Exact solution of some linear differential equations by Laplace transform.
International Journal of Nonlinear Science, 16, 3–11.
6. Khan, M., Hussain, M., Jafari, H., & Khan, Y. (2010). Application of Laplace decomposition
method to solve nonlinear coupled partial differential equations. World Applied Sciences
Journal, 9, 13–19.
7. Khelifa, S., & Cherruault, Y. (2000). New results for the Adomian method. Kybernetes, 29,
332–354.
8. Milici, C., & Drăgănescu, G. (2014). A method for solve the nonlinear fractional differential
equations. Saarbrücken: Lambert Academic Publishing.
9. Natanson, I. P. (1950). Teoria funcţii vescestvennoi peremennoi. Gosudarstvennoe izdatelstvo
tehniko-teoreticekoi literaturi, Moscva.
10. Pinkus, A. (2000). Weierstrass and approximation theory. Journal of Approximation Theory,
107, 1–66.
11. O’Shaughnessy, L. (1918). Problem 433. The American Mathematical Monthly, 25, 172.
12. Weilbeer, M. (2005). Efficient numerical methods for fractional differential equations and
their analytical background. PhD thesis, Facultät für Mathematik und Informatik, Technischen
Univerisität Braunschweig, Braunschweig.
www.pdfgrip.com
Chapter 5
Generalized Systems
This chapter addresses the generalization of classical models and systems in the
perspective of FC. The following sections study the Cornu, Emden, Hermite,
Legendre, and Bessel fractional systems.
5.1 Cornu Fractional System
5.1.1 Cos and Sin Fractional of Type Fresnel
We define fractional Cos, and Sin of order 0 < q ≤ 1, as:

t
1
C (t ) =
q 2
(t − u)q−1 cos u2 du,
(q) 0
t
1
S q (t 2 ) = (t − u)q−1 sin u2 du.
(q) 0
These functions can be represented in Mathematica with the aid of the following
program for q = 1/2 (see Fig. 5.1).
MATHEMATICA
Clear["‘*"]
q = 1/2;
u[t_] = 1/Gamma[q]*Integrate[(t-s)^(q-1)*Cos[s^2],
{s, 0, t}];

www.pdfgrip.com
88 5 Generalized Systems
1.0
0.5
0.5
0.0
0.0
–0.5
–0.5
–1.0
0 2 4 6 8 10 0 2 4 6 8 10
Fig. 5.1 The graph of fractional cosine (left) and sine (right) for q = 1/2
p = Plot[{u[t]}, {t, 0, 10}, PlotLabel -> "Cos

fractional",
ImageSize -> 200, Frame -> True];
v[t_] = 1/Gamma[q]*Integrate[(t-s)^(q-1)*Sin[s^2],
{s, 0, t}];
q = Plot[{v[t]}, {t, 0, 10}, PlotLabel -> "Sin
fractional",
ImageSize -> 200, Frame -> True];
Show[GraphicsArray[{p, q}]]
5.1.2 Cornu Fractional System and Curve
We will introduce the following Cornu1 fractional differential system with initial
conditions:
⎧ q
⎨ D x(t) = cos(t 2 ), x(0) = 0,
⎩
D q y(t) = sin(t 2 ), y(0) = 0.
1 M.A. Cornu (1841–1902).

www.pdfgrip.com
5.1 Cornu Fractional System 89
Solution We will solve this system using the LT. We have:

⎧
⎪ t4 t8 t 12
⎪
⎪ L [x (t)] = L cos(t ) = L 1 − + −
q 2
+ ... ,
⎪
⎨ 2! 4! 6!
⎪
⎪
⎪
⎪ t6 t 10 t 14
⎩ L [y q (t)] = L sin(t 2 ) = L t 2 − + − + ... ,
3! 5! 7!
⎧
⎪ 1 4! 8! 12!
⎪
⎨ s X = s − 2!s 5 + 4!s 9 − 6!s 13 + . . . ,
q
⎪
⎪
⎪
⎪
⎩ s q Y = 2! − 6! + 10! − 14! + . . . ,
s3 3!s 7 5!s 11 7!s 15
⎧
⎪ 1 4! 8! 12!
⎪
⎨ X = s q+1 − 2!s q+5 + 4!s q+9 − 6!s q+13 + . . . ,
⎪
⎪
⎪
⎪
⎩ Y = 2! − 6! + 10! − 14! + . . .
s q+3 3!s q+7 5!s q+11 7!s q+15
By inverse LT we obtain the solution {x(t), y(t)}:

⎧
⎪
⎪ tq 4! t q+4 8! t q+8 12! t q+12
⎪
⎪ x(t) = − + − + ...
⎨ (q + 1) 2! (q + 5) 4! (q + 9) 6! (q + 13)
⎪
⎪
⎪
⎪ 2t q+2 6! t q+6 10! t q+10 14! t q+14
⎩ y(t) = − + − + ...
(q + 3) 3! (q + 7) 5! (q + 11) 7! (q + 15)
The graph of this curve can be written in Maple and Mathematica as:
MAPLE
restart;a:=1/2:
> x(t):=1/GAMMA(a)*int((t-u)^(a-1)*cos(u^2),u=0..t):
> y(t):=1/GAMMA(a)*int((t-u)^(a-1)*sin(u^2),u=0..t):
> plot([x(t),y(t),t=0..10],color=black,
scaling=constrained);
MATHEMATICA
Clear["‘*"]
q = 1/2;
u[t_] = 1/Gamma[q]*Integrate[(t - s)^(q - 1)*Cos[s^2],
{s, 0, t}];
v[t_] = 1/Gamma[q]*Integrate[(t-s)^(q-1)*Sin[s^2],
{s, 0, t}];
www.pdfgrip.com
ParametricPlot[{u[t], v[t]}, {t, 0, 10},

Frame -> True, ImageSize -> 300]
Clear["‘*"]
Manipulate[
ParametricPlot[{1/Gamma[q]
*Integrate[(t - u)^(q - 1)*Cos[u^2], {u, 0, t}],
1/Gamma[q]*Integrate[(t - u)^(q - 1)*Sin[u^2],
{u, 0, t}]}, {t, 0, 10}, Frame -> True,
ImageSize -> 300], {{q, 3/2}, 0, 2}]
5.1.3 Cornu Generalized Curve/System
For 0 ≤ α ≤ 1 we can introduce a generalization of above curve and system, as:

⎧
⎪
⎪ α x(t) = cos tq
⎪
⎪ D , x(0) = 0
⎨ (q + 1)
⎪
⎪ tq
⎪
⎪
⎩ D α y(t) = sin , y(0) = 0
(q + 1)
5.1.4 Cornu Fractional System in a Plane
This curve can be plotted (see Fig. 5.2) in a plane using Maple and Mathematica as:
MAPLE
> restart;
> with(plots):
> a:=1/2:
> x:=1/GAMMA(a)*int((t-u)^(a-1)*cos(u^2),u=0..t):
> y:=1/GAMMA(a)*int((t-u)^(a-1)*sin(u^2),u=0..t):
> spacecurve([x,y,3 - x - y,t = 0..4],color=black,
scaling=constrained);
MATHEMATICA
Clear["‘*"]
Manipulate[
{x[t], y[t]} = {Integrate[Cos[u^2], {u, 0, t}],
Integrate[Sin[u^2], {u, 0, t}]};
p = ({x, y, z} /. First@Solve[x/a + y/b + z/c - 1
== 0, z]);
www.pdfgrip.com
a
b
c
1.0
0.5
1.0
0.0
–0.5
–1.0
0.0 0.5
0.5
0.0
1.0
Fig. 5.2 Cornu fractional curve in plane
q = p /. {x -> x[t], y -> y[t]};

Show[Plot3D[Evaluate[p[[3]]], {x, 0, 1},
{y, 0, 1}, Mesh->None,
PlotStyle -> {Blue, Opacity[0.05]}],
ParametricPlot3D[Evaluate[q], {t, 0, 2*Pi},
PlotStyle -> Blue],
ImageSize->300, AspectRatio->1], {{a, 1}, 0, 2},
{{b, 1}, 0, 2}, {{c, 1}, 0, 2}]
5.1.5 Fractional Cornu Spiral on the Sphere
Projection of the Cornu fractional curve on sphere (see Fig. 5.3) can be obtained in
Maple and Mathematica as:
www.pdfgrip.com
Fig. 5.3 Projection of the

Cornu fractional curve on
sphere
MAPLE
> restart;
> with(plots):
> a:=1/2:
> x:=1/GAMMA(a)*int((t-u)^(a-1)*cos(u^2),u=0..t):
> y:=1/GAMMA(a)*int((t-u)^(a-1)*sin(u^2),u=0..t):
> spacecurve([cos(x)*sin(y),sin(x)*sin(y),cos(y),
t=0..4],
color=black,scaling=constrained);
MATHEMATICA
Clear["‘*"]
p = ParametricPlot3D[{Cos[u]*Sin[v], Sin[u]*Sin[v],
Cos[v]},
{u, 0, 2*Pi}, {v, 0, Pi}, PlotStyle->{Cyan,
Opacity[0.3]},
Mesh -> None, Axes -> False, Boxed -> False];
q = 1/2;
{s, 0, t}];
v[t_] = 1/Gamma[q]*Integrate[(t - s)^(q - 1)*Sin[s^2],
{s, 0, t}];
q1 = ParametricPlot3D[{Cos[u[t]]*Sin[v[t]],
Sin[u[t]]*Sin[v[t]], Cos[v[t]]}, {t, 0, 10},
PlotStyle -> {Red, Opacity[0.3]},
www.pdfgrip.com
Mesh -> None, Axes -> False, Boxed -> False];

Show[p, q1]
5.1.6 Fractional Cornu Spiral on the Cone
This projection (see Fig. 5.4) can be obtained with Mathematica as:
Clear["‘*"]
x[u_, v_] := u*Cos[v];
y[u_, v_] := u*Sin[v];
z[u_, v_] := -u;
Manipulate[
{s, 0, t}];
v[t_] = 1/Gamma[q]*Integrate[(t - s)^(q - 1)*Sin[s^2],
{s, 0, t}];
p = ParametricPlot3D[{x[u, v], y[u, v], z[u, v]}, {
u, -1, 1}, {v, 0, 2*Pi}, Mesh -> None, Axes
-> False, Boxed -> False];
q1 = ParametricPlot3D[{x[u[t], v[t]], y[u[t], v[t]],
z[u[t], v[t]]}, {t, 0, 10}, Mesh -> None,
Axes -> False, Boxed -> False];
Show[p, q1], {{q, 1/2}, 0, 2}]
Fig. 5.4 Projection of the

Cornu fractional spiral on the
cone
www.pdfgrip.com
5.2 Power Series
We will use here the power series method [3] to solve the FDE. We denote by:
t t
1 α
f (t)(dt) =
α
(t − u)α−1 f (u)du = I α f (t),
(α + 1) 0 (α + 1 0
the FI of order α of f (t). Then, the norm of order α on LL2 of f (t) can be defined:
[0,1]
1 1/2
1
I α f (t)αL2 = f 2 (t)(dt)α ) .
[0,1] (α + 1) 0
5.2.1 The Müntz Theorem
For details, regarding this theorem,2 the reader can see [1, 7, 8].
We denote by C[a, b] the set of all continuous functions on [a, b].
Definition 4 For a function f ∈ C[a, b] the norm is defined
f = max |f (t)|,
t∈[a,b]
and for f, g ∈ C[a, b] it can be defined the distance:
f − g = max |f (t) − g(t)|.

t∈[a,b]
Definition 5 A sequence fn (t) of functions in C[a, b] converges uniformly to a

function f (t) if lim fn (t) − f (t) = 0.
n→∞
Definition 6 Let X be a metric space. If A ⊂ X then
"
A=A {lim an : ∀n > 0, an ∈ A},
n
is dense in A if:
A = X.
Theorem 7 The system
() = span{t α0 , t α1 , . . . , t αn , . . .}, αk ∈ R,
2 H. Müntz (1884–1956).
www.pdfgrip.com
5.2 Power Series 95
where
= {0 = α0 < α1 < α2 < . . .},
is dense in C[0, 1], if:

∞
1
= ∞.
αn
n=1
Proof We will establish the expression of error in L2 [0, 1]. We have:
t q − p(t)C[0,1] < qt q−1 − p (t)L2 [0,1] ,
p(t) ∈ (),
1
Er(t q , p(t))L2 [0,1]) = I [C1 , C2 , . . . , Cn ] = [t q − p(t)]2 dt
0
2
1
n
= tq − Ck t αk dt → min,
0 k=1
where t ∈ [0, 1], p(t) ∈ span{1, t α1 , t α2 , . . . , }, we have the functional:
2
1
n
I [C1 , C2 , . . . , Cn ] = t − q
Ck t αk
dt → min .
0 k=1
which, by minimization, give us the following system of equations in {C1 , C2 ,

. . . , Cn }:
∂I [C1 , C2 , . . . , Cn ]
= 0, i = 1, 2, . . . n.
∂Ci
This system can be written in an explicit form as:

⎧ 1 1 1 1
⎪
⎪ C1 + C2 + . . . + Cn = ,
⎪
⎪ α1 + α1 + 1 α1 + α2 + 1 α1 + αn + 1 α1 + q + 1
⎪
⎪
⎪
⎨ 1 1 1 1
C1 + C2 + . . . + Cn = ,
α2 + α1 + 1 α2 + α2 + 1 α2 + αn + 1 α2 + q + 1
⎪
⎪
⎪
⎪ ... ... ...
⎪
⎪
⎪
⎩ C1
1
+ C2
1
+ . . . + Cn
1
=
1
.
αn + α1 + 1 αn + α2 + 1 αn + αn + 1 αn + q + 1
www.pdfgrip.com
In the above system it appears the symmetric matrix:

⎛ ⎞
1 1 1
...
⎜ α1 + α1 + 1 α1 + α2 + 1 α1 + αn + 1 ⎟
⎜ ⎟
⎜ 1 1 1 ⎟
⎜ ... ⎟
H (α1 , α2 , . . . , αn ) = ⎜ α2 + α1 + 1 α2 + α2 + 1 α2 + αn + 1 ⎟ ,
⎜ ⎟
⎜ ... ... ... ⎟
⎝ 1 1 1 ⎠
...
αn + α1 + 1 αn + α2 + 1 αn + αn + 1
having the Gram3 determinant:
G(α1 , α2 , . . . , αn ) = det(H (α1 , α2 , . . . , αn )),
where Determinant (det) denotes determinant.

It is well known that:
)
G(q, α1 , α2 , . . . , αn )
Er(t , p(t))L2 [0,1] =
q
,
G(α1 , α2 , . . . , αn )
1
n
|q − αk |
Er(t q , p(t))L2 [0,1] = √ .
2q + 1 k=1 q + αk + 1
By recurrence, we can build the following set of functions:
Q0 (t) = t q ,
...
1
Qn (t) = (αn − q)t αn Qn−1 (u)u−(1+αn ) du.
t
After calculations, we obtain:

1
Q1 = (α1 − q)t α1
u−1−α1 uq du = t q − t α1 ,
t

n−1
Qn (t) = t q − Ck t αk .
k=0
3 J.P. Gram (1850–1916).

www.pdfgrip.com
5.2 Power Series 97
The last function can be verified by induction

1
Qn (t) = (αn − q)t αn
u−(1+αn ) Qn−1 (u) du,
t

1
n−2
Qn (t) = (αn − q)t αn u−(1+αn ) uq − Ck uαk du,
t k=0

n−2
Ck
Qn (t) = t q − t αn + (αn − q) (t αk − t αn ),
αn − αk
k=0
or, using the inequality
αt α (1 − t) < 1, where: t ∈ [0, 1], α > 0.
If α = 0, and t ∈ [0, 1], we obtain:

q q

Qn (t) ≤ Qn−1 1 − (1 − α n ) ≤ Qn−1 |1 − ,
αn αn

q
Qn (t)C[0,1] ≤ 1 − Qn−1 C[0,1] , n = 2, 3, . . . ,
αn
or

1 |αn − q| q
|αn − q|t αn
u−1−αn du = (1 − t αn ) ≤ 1 − ,
t αn αn
Q0 (t) = 1,
n

q
Qn (t) ≤ 1 − ,
αk
k=1

n q

ln Qn (t) = ln 1 − .
αk
k=1
But, for αk → ∞, we obtain:

q
q
ln 1 − ≈ − ,
αk αk
ln Qn (t) → −∞ ⇒ Qn (t) → 0.
Thus for all q ∈ N, Er(t q , p(t))C[0,1]) converges to zero as n → ∞.

www.pdfgrip.com
Remark We consider the functional

2
1 1
n
I= t − q
Ck t αk
(dt)α → min,
(α + 1) 0 k=0
and noting that

1 (αi + αj + 1)
t αi t αj (dt)α =
0 (αi + αj + α + 1)
we have the matrix (see Theorem 7):

⎛
(α1 + α1 + 1) (α1 + αn + 1) ⎞
...
⎜ (α1 + α1 + 1 + α) (α1 + αn + 1 + α) ⎟
⎜ ⎟
H (α1 , α2 , . . . , αn ) = ⎜ ... ⎟.
⎝ (α + α + 1) (αn + αn + 1) ⎠
n 1
...
(αn + α1 + 1 + α) (αn + αn + 1 + α)
We cannot calculate C1 , C2 , . . . , Cn and fractional error Er(t q , p(t)), in

L2α [0, 1]).
Example To approximate the function
f (t) = t 1/3 , t ∈ [0, 1],
using the function system:
{t 1/5 , t 1/2 , t 3/4 }.
Solution We will introduce the matrix A = [aij ], where

1
aij = t a t b dt, a, b ∈ {1/5, 1/2, 3/4} i, j = 1 . . . 3,
0
in order to write the following Maple application for calculation of the aij elements.
MAPLE
with(Student[LinearAlgebra]):
Digits:=4:
A:= 5/7, 10/17, 20/39 > | < 10/17, 1/2, 4/9 > | < 20/39, 4/9, 2/5 ;
b:= 15/23 > | < 6/11 > | < 12/25 ;
C:= evalf(LeastSquares(A,b));
f(t):=tˆ(1/3);
g(t):=0.3335*tˆ(1/5) + 0.9676*tˆ(1/2) - 0.3027*tˆ(3/4);
plot(f(t),g(t),t=0..1);
www.pdfgrip.com
5.2 Power Series 99
G:=abs(f(t) - g(t));
for t from 0 to 1 by 0.2 do G od;
It results:
C1 = 0.3335, C2 = 0.9676, C3 = −0.3027
|f (0) − g(0)| = 0, |f (0.2) − g(0.2)| = 0.00094
|f (0.4) − g(0.4)| = 0.0005, . . . , |f (1) − g(1)| = 0.001
Error of Type α
Matrix A = (aij )1≤i,j ≤3

1 (2/5 + 1)
a11 = t 2/5 (dt)α =
0 (2/5 + 1 + α)
1 (1/5 + 1/2 + 1)
a12 = t 1/5+1/2 (dt)α =
0 (1/5 + 1/2 + 1 + α)
1 (1/5 + 3/4 + 1)
a13 = t 1/5+3/4 (dt)α =
0 (1/5 + 3/4 + 1 + α)
a21 = a12
1 (2)
a22 = t (dt)α =
0 (2 + α)
1 (1/2 + 3/4 + 1)
a23 = t 1/2+3/4 (dt)α =
0 (1/2 + 3/4 + 1 + α)
a31 = a13
a32 = a23
1 (3/2 + 1)
a33 = t 3/2 (dt)α =
0 (3/2 + 1 + α)
1 (1/3 + 1/5 + 1)
b1 = t 1/3+1/5 (dt)α =
0 (1/3 + 1/5 + 1 + α)
1 (1/3 + 1/2 + 1)
b2 = t 1/3+1/2 (dt)α =
0 (1/3 + 1/2 + 1 + α)
www.pdfgrip.com
1 (1/3 + 3/4 + 1)
b3 = t 1/3+3/4 (dt)α =
0 (1/3 + 3/4 + 1 + α)
For b = (b1 , b2 , b3 ), C = (C1 , C2 , C3 ) we have
AC T = bT ⇒ C T = A−1 bT
MAPLE
restart;
Digits: = 4;
a11:= evalf(GAMMA(2/5 + 1)/GAMMA(2/5 + 1 + a));
a12:= evalf(GAMMA(1/5 + 1/2 + 1)/GAMMA(1/5 + 1/2 + 1 + a);
a13:= evalf(GAMMA(1/5 + 3/4 + 1)/GAMMA(1/5 + 3/4 + 1 + a);
a21:= a12;
a22:= evalf(A(2 + a))
a23:= evalf(1/2 + 3/4 + 1)/GAMMA(1/2 + 3/4 + 1 + a))
a31:= a13;
a32:= a23;
a33:= evalf(3/2 + 1)/GAMMA(3/2 +1 + a))
b1:= evalf(1/3 + 1/5 + 1)/GAMMA(1/3 + 1/5 + 1 + a))
b2:= evalf(1/3 + 1/2 + 1)/GAMMA(1/3 + 1/2 + 1 + a))
b3:= evalf(1/3 + 3/4 + 1)/GAMMA(1/3 + 3/4 + 1 + a))
ec1:= C1*a11 + C2*a12 + C3*a13 = b1;
ec2:= C1*a21 + C2*a22 + C3*a23 = b2;
ec3:= C1*a31 + C2*a32 + C3*a33 = b3;
solve({ec1,ec2,ec3},{C1,C2,C3});
for α = 2/3, C1 = 0.5444, C2 = 0.3366, C3 = 0.1264
f(t):= tˆ(1/3);
g(t):= 0.54448*tˆ(1/5) + 0.3366*tˆ(1/2) + 0.1264*tˆ(3/4);
plot(f(t),g(t),t = 0..1);
G:= abs(f(t) - g(t));
for t from 0 to 1 by 0.2 do G od;
From the last program, we obtain:
|f (0) − g(0)| = 0, |f (0.2) − g(0.2)| = 0.001,
|f (0.4) − g(0.4)| = 0.007, |f (0.6) − g(0.6)| = 0.005,
|f (0.8) − g(0.8)| = 0.004 |f (1) − g(1)| = 0.007.
Alternatively, in Mathematica we have:

MATHEMATICA
a11= Gamma(2/5 + 1)/Gamma(2/5 + 1 + a)//N
a12= Gamma(1/5 + 1/2 + 1)/Gamma(1/5 + 1/2 + 1 + a)//N
www.pdfgrip.com
5.2 Power Series 101
a13= Gamma(1/5 + 3/4 + 1)/Gamma(1/5 + 3/4 + 1 + a)//N

a21= a12;
a22= 1/Gamma(2 + a)//N
a23= Gamma(1/2 + 3/4 + 1)/Gamma(1/2 + 3/4 + 1 + a)//N
a31= a13
a32= a23
a33= Gamma(3/2 + 1)/Gamma(3/2 +1 + a)//N
b1= Gamma(1/3 + 1/5 + 1)/Gamma(1/3 + 1/5 + 1 + a)//N
b2= Gamma(1/3 + 1/2 + 1)/Gamma(1/3 + 1/2 + 1 + a)//N
b3= Gamma + 3/4 + 1)/Gamma + 3/4 + 1 + a)//N
ec1= C1*a11 + C2*a12 + C3*a13
ec2= C1*a21 + C2*a22 + C3*a23
ec3= C1*a31 + C2*a32 + C3*a33
Solve[{ec1,ec2,ec3}== {b1,b2,b3},{C1,C2,C3}]//N
f[t]= tˆ(1/3)
g[t]= 0.54448*tˆ(1/5) + 0.3366*tˆ(1/2) + 0.1264*tˆ(3/4)
Plot[{f[t],g[t]},{t,0,1}]
∞

Theorem 8 If Ck t kα is convergent for t = t0 , then it is convergent whenever
k=0
0 ≤ t < t0 .
∞

Proof Suppose that Ck t0kα is convergent.
k=0
Then sequence {ak t0kα } → 0, for k → ∞.
Thus, there is a constant M > 0, so that:

kα
Ck t0 ≤ M, k = 0, 1, . . .
Then
t kα
k k t
Ck t = Ck t0 ≤ M .
t0 t0
Again, if 0 ≤ t < t0 , then

t kα

< 1,
t0
∞
t kα
so is a convergent series.
t0
k=0
∞

Applying the comparison test, the series |Ck t kα | is convergent.
k=0
www.pdfgrip.com
∞

Then Ck t kα is absolutely convergent and therefore convergent.
k=0
∞

Remark If Ck t kα , diverges for t = t0 , then it diverges for t > t0 .
k=0
Example 1 Let us determine the solution of the FDE:
D (2α) y(t) + ω2 y = 0, 0 < α ≤ 1, t > 0,
y(0) = A, D (α) y(0) = B,
where A, B are constants. We apply now the series of powers method.
Solution We can write the solution as the series:

∞

y(t) = Cn t nα .
n=0
We have:
∞
∞
(nα + 1)
D (2α) y(t) = Cn D (2α) (t nα ) = Cn t (n−2)α
((n − 2)α + 1)
n=0 n=2
∞
((n + 2)α + 1) nα
= Cn+2 t .
(nα + 1)
n=0
Replacing in the equation, we have:

∞
∞
((n + 2)α + 1) 2nα
Cn+2 t + ω2 Cn t nα = 0.
(nα + 1)
n=0 n+0
Hence:
ω2 (nα + 1)
Cn+2 = − Cn , n = 0, 1, . . . ,
((n + 2)α + 1)
B ω2 A
C0 = A, C1 = , C2 = − , ...
(α + 1) (2α + 1)
∞
∞

(−1)n ω2n 2nα (−1)n ω2n
y(t) = A t +B t (2n+1)α ,
(2nα + 1) ((2n + 1)α + 1)
n=0 n=0
www.pdfgrip.com
or:
∞
(−1)n ω2n
y(t) = AE2α (−ω2 t 2α ) + B t (2n+1)α .
((2n + 1)α + 1)
n=0
Example 2 Let us determine the solution of the FDE [2]:
D (α) y(t) − y 2 − 1 = 0, m − 1 < α ≤ m, t > 0,
with the initial conditions
y (i) (0) = 0, i = 0, 1, . . . , m − 1,
using the series of powers method.

Solution We take the solution as a series of powers
∞

y(t) = Cn t nα .
n=0
We define the so-called αk-th FD as
D (kα) (D (α) y(t) − y 2 (t) − 1) = 0, k = 0, 1, . . . , m − 1,
and introducing the solution in the equation we obtain

∞ ∞

D (α(k+1)) Cn t nα − D (αk) Cn t nα = D (αk) (1), k = 0, 1, . . . ,
n=0 n=0
⎛ ⎞
∞
∞

n
(nα + 1) ⎝
Cn t (n−k−1)α − Cj Cn−j ⎠
((n − k − 1)α + 1)
n=k+1 n=k j =0
(nα + 1)
× t (n−k)α = χk ,
((n − k − 1)α + 1)
where
⎧
⎨ 1, k = 0,
χk =
⎩
0, k ≥ 1.
www.pdfgrip.com
By identification, we get the coefficients
C0 = 0,
1
C1 = ,
(α + 1)
(kα + 1) n
Ck+1 = Cj Ck−j ,
((1 + k)α + 1)
j =0
and the solution:

1 (2α + 1)
y(t) = tα + 2 t 3α + . . .
(α + 1) (α + 1)(3α + 1)
5.2.2 Lane-Emden Equation
We apply now the series of powers method to solve the FDE of Lane4 and Emden5
[5]
a1 a2 an
D α y(t) + D β1 y(t) + α−β D β2 y(t) + . . . + α−β D βn y(t) + y m (t) = 0,
t α−β1 t 2 t n
and the initial conditions are:
y(0) = 1, y (0) = 0.
We consider that 0 < t ≤ 1, 0 < βi ≤ 1, i = 1, 2, . . . , n, 1 < α ≤ 2, ai ∈ R,

and m ∈ N+ . The equation can be written also as Ly + y m = 0, using the linear
operator L.
Solution We take the approximate solution as a series of powers:
∞

y(t) = Ck t kα ,
k=0
where Ck are constants. The term Ly can be written also as:

n
ai
L[y(t)] = D α y + α−β
D βi y(t).
t i
i=0
4 J.H. Lane (1819–1880).

5 J.R. Emden (1862–1940).
www.pdfgrip.com
We will use also the formula:

(λ + 1) λ−α
Dα t λ = t .
(λ + 1 − α)
It results:
∞
∞
(αk + 1) kα−α (αk + 1 + α) kα
Dα y = Ck t = Ck+1 t
(αk + 1 − α) (αk + 1)
k=1 k=0
∞

ai ai (kα + 1)
D βi y(t) = Ck t kα−βi
t α−βi t α−βi (kα + 1 − βi )
k=1
∞
(kα + 1 + α)
= Ck+1 ai t kα .
(kα + 1 + α − βi )
k=0
The equation becomes:
∞

1
n
ai
Ck+1 (αk + 1 + α) + t αk + y m = 0.
(αk + 1) (kα + 1 + α − βi )
k=0 i=1
We introduce now the notation F (α, k):

1
n
ai
F (α, k) = Ck+1 (αk + 1 + α) + .
(αk + 1) (kα + 1 + α − βi )
i=1
We will examine some cases:

(a) In the case m = 0, we have:
∞

F (αk)t αk + 1 = 0.
k=0
Then:
C0 = 1,

n
ai
C1 (α + 1) 1 + + 1 = 0,
(1 + α − βi )
i=1
Ci = 0, i = 2, 3, . . .
www.pdfgrip.com
For α = 2, β = 1, a1 = 2, ai = 0, i = 2, 3, . . .
2
y (t) + y + 1 = 0,
t
resulting the solution:
t2
y(t) = 1 − .
6
(b) The case m > 0.
Then we have the following relations for different values of m:
1. m = 1, F (α, k) + Ck = 0,
k
2. m = 2, F (α, k) + Ci Ck−i = 0, because:
i=0
∞
2 ∞
k
y (t) =
2
Ck t αk
= Ci Ck−i t αk .
k=0 k=0 i=0

k
k−i
3. m = 3, F (α, k) + Ci Cj Ck−i−j = 0, because:
i=0 j =0
∞ 3 ∞
k
k−i
y (t) =
3
Ck t αk
= Ci Cj Ck−i−j t αk ,
k=0 k=0 i=0 j =0
or, the relation:
k−i−j −...tm−1

k
k−i
F (α, k) + ... Ci Cj . . . Ck−i−...−tm = 0.
i=0 j =0 tm =0
For same special cases, we have:
α = 2, β = 1, a1 = 2, ai = 0, i = 2, 3 . . . ,
2 t2
(a) m = 1, y (t) + y (t) + y(t) = 0 ⇔ y(t) = 1 − + . . . ,
t 6
2 t2 t4 sin t
(b) m = 2, y (t) + y (t) + y 2 (t) = 0 ⇔ y(t) = 1 − + + . . . = ,
t 3! 5! t
www.pdfgrip.com
2 t2 t4
(c) m = 3, y (t) + y (t) + y 3 (t) = 0 ⇔ y(t) = 1 − + + ...
t 3! 60
These results are confirmed by the following Maple program:
MAPLE
>m:= ... is introduced here
>ec:= diff(y(t),t,t) + 2/t*diff(y(t),t)+(y(t))^m = 0;
>dsolve({ec, y(0) = 1, D(y)(0) = 0}, y(t), series);
Other results are as follows:
t2 t4
(1) m = 4, y(t) = 1 − + + O(t 6 ),
6 30
t2 t4
(2) m = 5, y(t) = 1 − + + O(t 6 ),
6 24
t2 t4
(3) m = 6, y(t) = 1 − + + O(t 6 ),
6 20
t2 t4
(4) m = 10, y(t) = 1 − + + O(t 6 ),
6 12
t2 t4
(5) m = 20, y(t) = 1 − + + O(t 6 ).
6 6
Finally we will examine the following particular case:
3 2 1 1 3
y ( 2 ) (t) + y ( 2 ) (t) + 3 y ( 4 ) (t) + y m (t) = 0,
t t4
3 1 3
α= , β1 = , β2 = , a1 = 2, a2 = 1,
2 2 4
⎡

3 3 5 ⎢
⎢ 1
G(k) = F ,k = k+
2 2 2 ⎣ 3
k+1
2
⎤
2 1 ⎥
+ + ⎥ ,
3 3 3 ⎦
k+2 k+
2 2 4
G(0) = 5.07282, G(1) = 13.4199, G(2) = 24.9119, . . .

www.pdfgrip.com
∞
3
(a) m = 0, G(k)t 2 k + 1 = 0
k=0
1
C0 = 1, C1 = − = −0.19742, Ci = 0, i = 2, 3, . . . ,
G(0)
3
y(t) = 1 − 0.1971 t 2 ,
(b) m = 1, Ck+1 G(k) + Ck = 0,
1
C0 = 1, , C1 = − = −0.19742,
G(0)
C2 G(1) + C1 = 0 ⇒ C2 = 0.0146, . . . ,
3
y(t) = 1 − 0.1971t 2 + 0.0146t 3 + . . . ,

k
(c) m = 2, Ck+1 G(k) + Ci Ck−i = 0,
i=0
C0 = 1,
C1 G(0) + C02 = 0 ⇒ C1 = −0.1971,
C2 G(1) + 2C0 C1 = 0 ⇒ C2 = 0.0293,

3
y(t) = 1 − 0.1971t 2 + 0.0293t 3 + . . .
5.2.3 The Taylor Series Method
∞

Theorem 9 The series Ck t kα , here C0 = 0, is convergent for 0 < t < R 1/α .
k=0
Proof We can write:
C t (n+1)α
n+1
< 1,
Cn t nα
or:
C
n
0 < tα < .
Cn+1
www.pdfgrip.com
C
n
But the convergence radius is R = lim so that, finally, we obtain:
n→∞ Cn+1
0 < t < R 1/α .
Theorem 10 Suppose that:

∞

f (t) = Cn t nα , 0 ≤ m − 1 < α ≤ m, 0 ≤ t < R 1/α .
n=0
If f (t) ∈ C[0, R 1/α ), and D (nα) f (t) ∈ C(0, R 1/α ), for n = 0, 1, . . ., then:
D nα f (0)
Cn = ,
(nα + 1)
where D (nα) = D (α) D (α) . . . D (α) = (D (α) )n .

Proof For t = 0, we have C0 = f (0).
Using the formula
(λ + 1)
D (nα) t λ = t λ−nα ,
(λ + 1 − nα)
it results:
(2α + 1) α (3α + 1) 2α
D (α) f (t) = C1 (α + 1) + C2 t + C3 t + ...
(α + 1) (2α + 1)
For:
D (α) f (0)
t = 0, we have: C1 = .
(α + 1)
However, if we continue to apply n times D (α) we obtain for t = 0:
D (nα) f (0)
Cn = .
(nα + 1)
Thus we have the generalized MacLaurin series:

∞
D (nα) f (0)
f (t) = t nα .
(nα + 1)
n=0
www.pdfgrip.com
Remark Similarly, for t0 ≤ t < R 1/α and
∞

f (t) = (t − t0 )nα ,
n=0
we have the Taylor generalized formula:
∞
(nα)
Dt f (t0 )
f (t) = 0
(t − t0 )nα .
(nα + 1)
n=0
5.2.4 The Generalized Hermite Equation
The Hermite6 equation [6] can be generalized as [4]:
D (2α) y(t) − 2t α D (α) y(t) + λy(t) = 0,
for the fractional case.

Solution The solution can be written using the powers series method, as:
∞

y(t) = Cn t nα , (C0 = 0),
n=0
where Cn are constants.

In the calculations we apply the formula:
(λ + 1) λ−α
Dα t λ = t .
(λ + 1 − α)
It results:
∞
∞
∞

Cn D (2α) t nα − 2t α Cn D (α) t nα + λ Cn t nα = 0,
n=2 n=1 n=0
∞
∞
(nα + 1) (nα + 1)
Cn t (n−2)α −2t α Cn t (n−1)α
((n − 2)α + 1) ((n − 1)α + 1)
n=2 n=1
∞

+λ Cn t nα = 0,
n=0
6 C. Hermite (1822–1901).
www.pdfgrip.com
or, finally:
∞
∞ ∞
((n + 2)α + 1) nα (nα + 1)
Cn+2 t −2 Cn t nα + λ Cn t nα = 0
(nα + 1) ((n − 1)α + 1)
n=0 n=1 n=0
λC0
For n = 0 ⇒ C1 (2α + 1) + λC0 = 0 ⇒ C1 = −
(2α + 1)

((n + 2)α + 1) 2(nα + 1)
Cm+1 = Cn −λ .
(nα + 1) ((n − 1)α + 1)
For α = 1, we obtain:
λC0
C1 = − ,
2!
2n − λ
Cn+1 = Cn ,
(n + 1)(n + 2)

1 2 1
y(t) = C0 1 − λt + λ(λ − 4)t + . . . .
3
2! 4!
1
For α = , we have
2

2λC9 λ 2 C0 2
C1 = −λC0 , C2 = , C3 = √ − λ , ...,
3 3 π
and the solution:

2λ λ2 2
y(t) = C0 1 − λt 1/2 + t+ √ − λ t 3/2
3 3 π
5.2.5 The Generalized Legendre Equation
The generalized Legendre FDE can be defined as [4]:
(1 − t 2α )D (2α) y(t) − 2t α D (α) y(t) + λy(t) = 0.
Solution We consider a power series solution:

∞

y(t) = Cn t nα , C0 = 0
n=0
www.pdfgrip.com
where Cn are constants. We replace the solution in the equation

∞
∞
∞

(1 − t 2α ) Cn D (2α) t αn − 2t α Cn D (α) t αn + λ Cn t αn = 0,
n=2 n=1 n=0
or:
∞
∞
(nα + 1) (nα + 1)
Cn t (n−2)α − Cn t nα
((n − 2)α + 1) ((n − 2)α + 1)
n=2 n=2
∞
(nα + 1)
−2 Cn t nα
((n − 1)α + 1)
n=1
∞

+λ Cn t nα = 0,
n=0
and, finally we obtain:

∞
∞
((n + 2)α + 1) nα (nα + 1)
ds Cn+2 t − Cn t nα
(nα + 1) ((n − 2)α + 1)
n=0 n=2
∞
(nα + 1)
−2 Cn t nα
((n − 1)α + 1)
n=1
∞

+λ Cn t nα = 0.
n=0
For n = 0, we have:
C0 λ
C2 (2α + 1) + λC0 = 0, ⇒ C2 = − .
(2α + 1)
For n = 1, we obtain:
(3α + 1) λ − 2(α + 1)
C3 −2C1 (α +1)+λC1 = 0, ⇒ C3 = −C1 (α +1)
(α + 1) (3α + 1)
(α(n + 2) + 1) (αn + 1)
Cn+2 −Cn
(αn + 1) (α(n − 2) + 1)
(αn + 1)
−2Cn + λCn = 0.
(α(n − 1) + 1)
www.pdfgrip.com
For α = 1
λC0
C2 = − ,
6
λ−2
C3 = −C1 ,
6
λ(λ − 6)
C 4 = C0 ,
24
...
and the solution:

λ λ(λ − 6) 4 λ−2 3
y(t) = C0 1 − t 2 + t + . . . + C1 t − t + ...
2 24 6
1
For α = , we have:
2
λC0
C2 = − ,
6
√
λ− π
C3 = −2C1 ,
3

1 4
C4 = −C0 1−λ+ √ .
12 π
5.2.6 The Generalized Bessel Equation
The Bessel7 FDE can be introduced as [4]:
t 2α D (2α) y(t) + t α D (α) y(t) + (t 2α − p2 )y(t) = 0, p ∈ R.
Solution We consider a power series solution:

∞

y(t) = Cn t λ+nα , (C0 = 0),
n=0
7 F. Bessel (1784–1846).
www.pdfgrip.com
where Cn are constants. Replacing the solution in the equation, we have:

∞
(λ + nα + 1)
t 2α D (2α) y(t) = t 2α Cn t λ+nα−2α
(λ + (n − 2)α + 1)
n=0
∞
(λ + nα + 1)
= Cn t λ+nα ,
(λ + (n − 2)α + 1)
n=0
or
∞
(λ + nα + 1)
t α D (α) y(t) = t α Cn t λ+nα−α
(λ + (n − 1)α + 1)
n=0
∞
(λ + nα + 1)
= Cn t λ+nα ,
(λ + (n − 1)α + 1)
n=0
and, finally:
∞
∞
∞

(t 2α
−p ) 2
Cn t λ+nα
= Cn t λ+(n+2)α
−p 2
Cn t λ+nα
n=0 n=0 n=0
∞
∞

= Cn−2 t λ+nα − p2 Cn t λ+nα .
n=0 n=0
Hence, the recurrence relation between the coefficients Cn is:

∞
∞
(λ + nα + 1) λ+nα (λ + nα + 1) λ+nα
Cn t + Cn t
((n − 2)α + 1) ((n − 1)α + 1)
n=0 n=0
∞
∞

+ Cn−2 t λ+nα − p2 Cn t λ+nα = 0.
n=0 n=0
From the recurrence relation, it results:

(λ + 1) (λ + 1)
C0 + − p2 = 0,
(λ − 2α + 1) (λ − α + 1)

(λ + α + 1) (λ + α + 1
C1 + − p2 = 0,
(λ − α + 1) (λ + 1)
...,

(λ + kα + 1 (λ + kα + 1)
Ck + − p + Ck−2 = 0.
2
(λ + (k − 2)α + 1) (λ + 1)
www.pdfgrip.com
For α = 1, we have:

C0 λ2 − p2 = 0,

C1 (λ + 1)2 − p2 = 0.
For C0 arbitrary, λ = ±p, (λ + 1 = 0) results C1 = 0.
C2k−2
C2k = − ,
(2p + 2k)(2k)
or:
C0
C2k = (−1)k+1 .
2 · 4 · 6 · · · 2k (2p + 2) · · · (2p + 2k)
We obtain the solution:

∞
(−1)k t p+2k
y(t) = C0 ,
4k k!(p + 1)(p + 2) · · · (p + k)
k=0
and
∞
1 (−1)k t p+2k
Jp (t) = ,
2p (p + 1) 4k k!(p + 1)(p + 2) · · · (p + k)
k=0
∞
p+2k
(−1)k t
Jp (t) = ,
k!(p + k + 1) 2
k=0
where Bessel function of first kind (Jp (t)) is called the Bessel function of first kind.
1
For α = , C0 arbitrary, C1 = 0, using Mathematica we will solve the equation:
2
λ(λ)
λ+ − p2 = 0.
1
λ+
2
For p = 2, if we apply the Mathematica command:
F indRoot[x + x ∗ [x + 1/2] − 4 == 0, {x, 0.1}],
we obtain x = λ = 2.37593, [3].

www.pdfgrip.com
5.2.7 Nonlinear Systems

Lotka System
We consider as a first example the Lotka8 system with initial conditions:

⎧ α
⎨ D x(t) = 3.5 y(t) (1 − y(t)), x(0) = 0.2,
⎩
D α y(t) = 4 x(t) (1 − x(t)), y(0) = 0.2.
We look here for solutions in the form of series of powers:

∞
∞

x(t) = an t nα
, y(t) = bn t nα .
n=0 n=0
We use the derivation rule:

(λ + 1) λ−α
Dα t λ = t .
(λ + 1 − α)
It results:
∞ ∞ ∞
(nα + 1)
D α
an t nα
= an D α t nα =
((n − 1)α + 1)
n=0 n=1 n=1
∞
∞
((m + 1)α + 1) mα ((n + 1)α + 1) nα
= t = t .
(mα + 1) (nα + 1)
m=0 n=0
We obtain the following recurrence relations:

⎧
⎪
⎪ ((n + 1)α + 1)
n−k
⎪
⎪ an+1 = 3.5(bn − bk bn−k ),
⎪
⎪ (nα + 1)
⎪
⎪
⎪
⎪
k
⎪
⎨ a0 = 0.2,
⎪
⎪
⎪
⎪ ((n + 1)α + 1)
n−k
⎪
⎪ bn+1 = 4 an −
⎪
⎪ ak an−k ,
⎪
⎪ (nα + 1)
⎪
⎩ k
b0 = 0.2.
8 A.J. Lotka (1880–1949).

www.pdfgrip.com
For numerical calculations of an , bn , we can use the Mathematica program:

MATHEMATICA
Clear["‘*"]
\[Alpha] := 3/4;
a[0] = 0.1; b[0] = 0;
f[n_] := Gamma[n*\[Alpha] + 1]/Gamma[(n + 1)*
\[Alpha]+1];
For[n = 0, n <= 5,
n++, {a[n + 1] = 3.5*f[n]*(b[n] -
Sum[b[k]*b[n - k], {k, 0, n}]),
b[n + 1] = 4*f[n]*(a[n] -
Sum[a[k]*a[n - k], {k, 0, n}]), {n, 0, 5}}]
TableForm[Table[{n, a[n], b[n]}, {n, 0, 5}]]
0 0.1 0
1 0 0.3917
2 0.9475 1.5816
3 -0.2800 1.58164
4 2.3520 -0.3807
5 -2.0508 1.42210
sol = Table[{a[n], b[n]}, {n, 1, 1000}];

p = Interpolation /@ Transpose@sol;
ParametricPlot[Evaluate@Through@p@t,
{t, 1, 1000}, Frame -> True]
Figure 5.5 shows the solution (a(t), b(t)) of the Lotka attractor.
Fig. 5.5 Lotka attractor

solution (a(t), b(t))
1.0
0.5
0.0
–2 –1 0 1 2
www.pdfgrip.com
Lorenz Fractional Attractor
The Lorenz9 fractional attractor is defined by the system of FDE:

⎧
⎪
⎪ D α x(t) = 10 (y(t) − x(t)),
⎪
⎪
⎪
⎪
⎨ α
D y(t) = x(t) (28 − z(t)) − y(t),
⎪
⎪
⎪
⎪
⎪
⎪
⎩ D α z(t) = x(t) y(t) − 8 z(t).
3
We use the initial conditions:
x(0) = 0.1, y(0) = 0.1, z(0) = 0.1.
We take the power series solutions and 0 < α ≤ 1:

∞
∞
∞

x(t) = an t nα , y(t) = bn t nα , z(t) = cn t nα .
n=0 n=0 n=0
We obtain the recurrence relations:

⎧
⎪ ((n + 1)α + 1)
⎪
⎪ an+1 = 10[bn − an ]
⎪
⎪ (nα + 1)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ ((n + 1)α + 1) n
bn+1 = 28an − ak cn−k − bn
⎪ (nα + 1)
⎪
⎪ k=0
⎪
⎪
⎪
⎪
⎪
⎪ ((n + 1)α + 1)
n
⎪
⎪ 8
⎪ n+1
⎩ c = ak bn−k − cn .
(nα + 1) 3
k=0
Example For α = 0.95, we have:

⎧
⎪ D x(t) = 10[y(t) − x(t)]
⎪ 0.95
⎪
⎪
⎪
⎪
⎨ 0.95
D y(t) = x(t)[28 − z(t)] − y(t)
⎪
⎪
⎪
⎪
⎪
⎪
⎩ D 0.95 z(t) = x(t)y(t) − 8 z(t)
3
x(0) = 0.1, y(0) = 0.1, z(0) = 0.1.
9 E.N. Lorenz (1917–2008).

www.pdfgrip.com
The coefficients an , bn , cn can be calculated using the program:

MAPLE
restart:Digits:=5:
> a[0]:=0.1:
> b[0]:=0.1:
> c[0]:=0.1:
> alpha:=0.95:
> unassign(’n’);
> f:=n->GAMMA(n*alpha+1)/GAMMA((n+1)*alpha+1);
GAMMA(n alpha + 1)
f := n -> ------------------------
GAMMA((n + 1) alpha + 1)
> a[1]:=f(0)*10*(b[0]-a[0]):
> b[1]:=f(0)*(28*a[0]-a[0]*c[0]-b[0]):
> c[1]:=f(0)*(a[0]*b[0]-8/3*c[0]):
> for n from 1 to 5 do
> a[n+1]:=f(n)*10*(b[n]-a[n]);
> b[n+1]:=f(n)*(28*a[n]-sum(a[k]*b[n-k],k=0..n)-b[n]);
> c[n+1]:=f(n)*(sum(a[k]*b[n-k],k=0..n)-8/3*c[n]);
> od;
The solutions are:
x(t) = 0.1 − 0t 0.95 + 14.720t 2·0.95 − 59.987t 3·0.95 + 519.94t 4·0.95 −
−2523.97t 5·0.95 + 12242t 6·0.95 . . .
y(t) = 0.1 + 2.7451t 0.95 − 1.6192t 2·0.95 + 151.18t 3·0.95 − 524.76t 4·0.95 +
+3900t 5·0.95 − 14988t 6·0.95 . . .
z(t) = 0.1 − 0.26193t 0.95 + 0.52174t 2·0.95 − 0.02977t 3·0.95 +
+13.868t 4·0.95 − 49.518t 5·0.95 + 802.5t 6·0.95 . . .
For α = 0.995, the coefficients an , bn , cn are:

MATHEMATICA
In this case, we have:
Clear["‘*"]
\[Alpha] := 0.995;
a[0] = 0.1; b[0] = 0.1; c[0] = 0.1;
www.pdfgrip.com
f[n_] := Gamma[n*\[Alpha] + 1]/Gamma[(n + 1)

*\[Alpha] + 1];
a[1] = f[0]*10*(b[0] - a[0]);
b[1] = f[0]*(28*a[0] - a[0]*c[0] - b[0]);
c[1] = f[0]*(a[0]*b[0] - 8/3*c[0]); For[n = 0, n <= 5,
n++, {a[n + 1] = 10*f[n]*(b[n] - a[n]),
b[n + 1] = f[n]*(28*a[n] - Sum[a[k]*b[n - k],
{k, 0, n}]),
c[n + 1] = f[n]*(Sum[a[k]*b[n - k], {k, 0, n}]
- 8/3*c[n])}]
TableForm[Table[{n, a[n], b[n], c[n]}, {n, 0, 5}]]
References
1. Almira, J. M. (2007). Müntz type theorem I. Surveys in Approximation Theory, 3, 152–194.

2. El-Ajou, A., Arqub, O. A., Al Zhour, Z., & Momani, S. (2013). Application of a variational
iteration method to linear and nonlinear viscoelastic models with fractional derivatives. Entropy,
15, 5305–5323.
3. Johnson, R. S. (2006). The notebook series. the series of second order ordinary differential
equation and special function. Technical report, School of Mathematics & Statistics, University
of Newcastle upon Tyne.
4. Milici, C., & Drăgănescu, G. (2017). Generalization of the equations of Hermite, Legendre and
Bessel for the fractional case. Journal of Applied Nonlinear Dynamics, 6, 243–249.
5. Milici, C., & Drăgănescu, G. (2017). The Lane-Emden fractional homogenous differential
equation. Journal of Applied Nonlinear Dynamics, 6, 237–242.
6. Nikiforov, A. F., & Ouvarov, V. (1976). Eléments de la théorie des fonctions spéciales. Moscow:
Mir Publishers.
7. Rudin, W. (1966). Fractional calculus with applications for nuclear reactor dynamics. New
York: McGraw-Hill.
8. von Golitschek, M. (1983). A short proof of Müntz theorem. Journal of Approximation Theory,
39, 394–395.
www.pdfgrip.com
Chapter 6
Numerical Methods
This chapter studies several numerical methods for fractional order systems. In
the following sections variational iteration, least squares, Euler’s, and Runge–Kutta
methods are analyzed.
6.1 Variational Iteration Method for Fractional Differential

Equations
In this section we will introduce the variational iteration method (Variational

iteration method (VIM)) [3, 4] to solve the nonlinear FDE [2, 5, 6]. We consider
the FDE:
D α y(t) + R[y(t)] + N [y(t)] = g(t),
where g(t) is a given function, N[y(t)] is a nonlinear operator, and R is a residual

linear operator. The equation will be solved for the initial conditions:
y (k) (0) = y0(k) , m − 1 < α ≤ α

= m, k = 0, . . . , m − 1,
considering that the conditions of existence and uniqueness are satisfied.

We apply the LT method:
L[y(t)] = Y (s) = Y,

m−1
(k)
L[D α y(t)] = s α Y − y0 s α−k−1 .
k=0

www.pdfgrip.com
122 6 Numerical Methods
We will introduce now the correction relation, written in a recurrent form:

m−1
(k)
Yn+1 = Yn + λ(s) s Yn − α
y0 s α−k−1 + L[R[yn ] + N [yn ] − g(t)] ,
k=0
where λ(s) is the Lagrange1 multiplier.

δYn+1
We impose the condition = 0. It follows:
δYn
1
1 + λ(s)s α = 0 ⇒ λ(s) = − .
sα
It results:
1 (k) α−k−1
m−1
1
Yn+1 = α y0 s − α L[R[yn ] + N [yn ] − g(t)],
s s
k=0
or:

1 (k) α−k−1
m−1
−1 1
yn+1 (t) = L y0 s − α L[R[yn ] + N [yn ] − g(t)] .
sα s
k=0
Example 1 Use the VIM method to solve the FDE:

t
D α y(t) = 1 + y(u) du,
0
y(0) = 1, 0 < α ≤ 1.
Solution Using LT we have successively:
L[y(t)] = Y,
t
1
s Y −s
α α−1
= +L y(u) du ,
s 0
t
1
Yn+1 = Yn + λ(s) s α Yn − s α−1 − − L y(u) du ,
s 0
1 Joseph-Louis Lagrange (1736–1813).

www.pdfgrip.com
6.1 Variational Iteration Method for Fractional Differential Equations 123
δYn+1 1
= 0 ⇒ 1 + λ(s)s α = 0 ⇒ λ(s) = − α ,
δYn s
t
−1 1 1 1
yn+1 = L + α+1 + α L y(u) du ,
s s s 0

tα −1 Yn
yn (t) = 1 + +L
(α + 1) s α+1
y0 = 1,
tα t α+1
y1 = 1 + + ,
(α + 1) (α + 2)
...
Example 2 Use the VIM to solve the FDE:

t
D α y(t) = 1 + (t − u)y(u) du,
0
y(0) = 1, y (0) = 0, 1 < α ≤ 2.
Solution Using LT we have:
L[y(t)] = Y,
1 1
s α Y − s α−1 = + 2 L [yn (u)] ,
s s

1
Yn+1 = Yn + λ(s) s α Yn − s α−1 − − L [yn ] ,
s
δYn+1 1
= 0 ⇒ 1 + λ(s)s α = 0 ⇒ λ(s) = − α ,
δYn s

1 1 1
yn+1 = L−1 + α+1 + α+2 L [yn ] ,
s s s

tα Yn
yn (t) = 1 + + L−1 α+2 ,
(α + 1) s
y0 = 1,
www.pdfgrip.com
tα t α+2
y1 = 1 + + .
(α + 1) (α + 3)
...
Other details on VIM can be found in [2–4].
6.2 The Least Squares Method
Consider the equation
D α y(t) + y(t) + f (t) = 0,
y(0) = 0, y(1) = 0, 0 < α ≤ 1,
using the approximation:

n
yapp = Ci φ i ,
i=1
where Ci are constants, and φi = φi (t), i = 1, . . . , n are test functions. For

calculations, we will introduce an operator L:
L[yapp ] = D α yapp + yapp .
We will define the functional I

1
1 2
I [C1 , C2 , . . . , Cn ] = L[yapp ] + f (t) (dt)α → min,
(α + 1) 0
which, by minimization, gives a system of equations in C1 , . . . , Cn :
∂I [C1 , C2 , . . . , Cn ]
= 0, i = 1, 2, . . . n.
∂Ci
from which we obtain the constants C1 , . . . , Cn .

Example 1 Establish an approximate solution, with the aid of least squares method,
for the FDE:
D α y(t) + 1 − (1 + α)t = 0,
www.pdfgrip.com
6.2 The Least Squares Method 125
y(0) = 0, y(1) = 0, 0 < α ≤ 1.
Solution Exact solution, using the LT method is:
tα t α+1
s α Y = −L[1] + (α + 1)L[t], ⇔ y(t) = − + .
(α + 1) (α + 1)
For
yapp = C1 φ1 + C2 φ2 ,
D α yapp = C1 D α φ1 + C2 D α φ2 .
we obtain:
1
1 2
I [C1 , C2 ] = C1 D α φ1 + C2 D α φ2 + 1 − (α + 1)t (dt)α → min,
(α + 1) 0
⎧ ∂I [C , C ]
⎪
⎪
1 2
= 0,
⎪
⎨ ∂C1
⎪
⎪
⎩ ∂I [C1 , C2 ] = 0,
⎪
∂C2
⎧ 1
⎪
⎪
⎪ C1 D α φ1 + C2 D α φ2 + 1 − (1 + α)t D α φ1 (dt)α = 0,
⎪
⎨ 0
⎪
⎪ 1
⎪
⎪
⎩ C1 D α φ1 + C2 D α φ2 + 1 − (1 + α)t D α φ2 (dt)α = 0,
0
and using the notation

1 1
Bij = (D α φi )(D α φj )(dt)α , Fi = − (1 − (α + 1)t)D α φi (dt)α ,
0 0
where i, j = 1, 2. It follows:
⎛ ⎞
B11 B12 ,
B=⎝ ⎠, C = [C1 , C2 ], F = [F1 , F2 ].
B21 B22 ,
www.pdfgrip.com
Finally, the system will be:
BC T = F T ,
C T = B −1 F T .
Using the formulas:
(k + 1) k−α
Dα t k = t
(k + 1 − α)
(k + 1) k+α
I αtk = t ,
(k + 1 + α)
We choose the test functions as:
v1 = t (1 − t), v2 = t 2 (1 − t),
yapp = C1 v1 + C2 v2 ,
D α yapp = C1 D α v1 + C2 D α v2 .
1 2
D α v1 = D α t (1 − t) = t 1−α − t 2−α ,
(2 − α) (3 − α)
2 6
D α v2 = D α (t 2 − t 3 ) = D α t 2 − D α t 3 = t 2−α − t 3−α .
(3 − α) (4 − α)
We introduce the notations:

1 2 6
P = , Q= , R= .
(2 − α) (3 − α) (4 − α)
We obtain:
D α v1 = P t 1−α − Q t 2−α ,
D α v2 = P t 2−α − R t 3−α ,
D α v1 D α v1 = P 2 t 2−2α − 2P Q t 3−2α + Q2 t 4−2α ,
D α v1 D α v2 = P Q t 3−2α − P R t 4−2α − Q2 t 4−2α + QR t 5−2α ,
D α v2 D α v2 = Q2 t 4−2α − 2QR t 5−2α + R 2 t 6−2α .

www.pdfgrip.com
Using the integration rule
(k + 1)
0I x =
α k
,
(k + 1 + α)
we obtain:
1 (3 − 2α) (4 − 2α) (5 − 2α)
B11 = D α v1D α v1(dt)α = P 2 − 2P Q + Q2 ,
0 (3 − α) (4 − α) (5 − α)
(5 − 2α) (5 − 2α) (5 − 2α) (6 − 2α)

B12 = B21 = P Q −PR − Q2 + QR ,
(5 − α) (5 − α) (5 − α) (6 − α)
(5 − 2α) (6 − 2α) (7 − 2α)

B22 = Q2 − 2QR + R2 ,
(5 − α) (6 − α) (7 − α)
1
F1 = − [1 − (α + 1)t][P t 1−α − Qt 2−α ](dt)α
0
(2 − α) (3 − α) (3 − α) (2 − α)
= −P +Q + P (α + 1) − Q(α + 1) ,
(2) (3) (3) (2)
1
F2 = − [1 − (α + 1)t][Qt 2−α − Rt 3−α ](dt)α
0
(3 − α) (4 − α) (4 − α) (5 − α)
= −Q +R + Q(α + 1) − R(α + 1) .
(3) (4) (4) (5)
We can build now the matrix B:

⎛ ⎞
B11 B12 ,
B=⎝ ⎠,
B21 B22 ,
C = [C1 C2 ],
F = [F1 F2 ],
and solve the matrix equation:
BC T = F T ⇒ C T = B −1 F T ,
resulting C1 and C2 .
The programs in Maple and Mathematica are listed in the follow-up.
www.pdfgrip.com
MAPLE
> restart:
> Digits:= 4:
> a:= 1/2;
> P:= evalf(1/GAMMA(2 - a));
> Q:= evalf(2/GAMMA(3 - a));
> R:= evalf(6/GAMMA(4 - a));
> B11:=evalf(P^2*GAMMA(3 - 2*a)/GAMMA(3 - a)
- 2*P*Q*GAMMA(4 - 2*a)/GAMMA(4-a) +
Q^2*GAMMA(5 - 2*a)/GAMMA(5 - a));
> B12:=evalf(P*Q*GAMMA(4 - 2*a)/GAMMA(4 - a)
-P*R*GAMMA(5 - 2*a)/GAMMA(5 - a) - Q^2
*GAMMA(5 - 2*a)
/GAMMA(5 - a) + Q*R*GAMMA(6 - 2*a)/GAMMA(6 - a));
> B22:=evalf(P^2*GAMMA(5 - 2*a)/GAMMA(5 - a) -
2*Q*R*GAMMA(6 - 2*a)/GAMMA(6 - a) + R^2
*GAMMA(7 - 2*a)
/GAMMA(7 - a) );
> F1:=evalf(2/GAMMA(3 + a) - 6/GAMMA(4 + a));
> F2:=evalf(6/GAMMA(4 + a) - 24/GAMMA(5 + a));
> ec1:= C1*B11 + C2*B12 = F1;
> ec2:= C1*B12 + C2*B22 = F2;
> solve({ec1,ec2},{C1,C2});
MATHEMATICA
Clear["‘*"]
a = 1/2
P = 1/Gamma[2 - a] // N
Q = 2/Gamma[3 - a] // N
C1 = 6/Gamma[4 - a] // N
v1 = P*t - B*t^2 // N
v2 = P*t^2 - C1*t^3 // N
B11 = P^2*Gamma[3 - 2*a]/Gamma[3 - a] -
2*P*Q*Gamma[4 - 2*a]/Gamma[4 - a] +
Q^2*Gamma[5 - 2*a]/Gamma[5 - a] // N
B12 = P*Q*Gamma[4 - 2*a]/Gamma[4 - a] -
P*C1*Gamma[5 - 2*a]/Gamma[5 - a] -
Q^2*Gamma[5 - 2*a]/Gamma[5 - a] +
Q*C1*Gamma[6 - 2*a]/Gamma[6 - a] // N
B22 = Q^2*Gamma[5 - 2*a]/Gamma[5 - a] -
2*Q*C1*Gamma[6 - 2*a]/Gamma[6 - a] +
C1^2*Gamma[7 - 2*a]/Gamma[7 - a] // N
F1 = -P*Gamma[2 - a]/Gamma[2] + Q*Gamma[3 -
a/Gamma[3]] +
www.pdfgrip.com
P*(a + 1)*Gamma[3 - a]/Gamma[3]

-Q*(a + 1)*Gamma[4 - a]/Gamma[4]
F2 = -Q*Gamma[3 - a]/Gamma[3] + C1*Gamma[4 - a]/
Gamma[4] +
Q*(a + 1)*Gamma[4 - a]/Gamma[4]
-C1*(a + 1)*Gamma[5 - a]/Gamma[5]
ec1 = C11*B11 + C2*B12
ec2 = C11*B12 + C2*B22
Solve[{ec1, ec2} == {F1, F2}, {C1, C2}] // N
For α = 0.50, we obtain:
C1 = 54.3332, C2 = −39.3793.
Example 2 Establish an approximate solution, with the aid of least squares method,
for the FDE
t
D α y(t) − y(t) = , 0 ≤ t ≤ 1,
et − 1
y(0) = 1, y (0) = 0, where: 0 < α ≤ 1.
Solution It is well known that f (t) can be expanded as:

∞
tk
f (t) = Bk ,
k!
k=0
where Bk are Bernoulli2 numbers. In our case:
1 1 1
B0 = 1, B1 = − , B2 = , B4 = − , . . . , B2k+1 = 0.
2 6 30
The approximate solution is in this case:
∞

yapp = Ck t αk ,
k=0
where Ck are constants.
2 J. Bernoulli (1655–1705).
www.pdfgrip.com
We apply now the least squares method. We introduce the functional:
∞
2
1 tk
I= D yapp − yapp −
α
Bk (dt)α → min,
0 k!
k=0
which will be minimized. We can calculate also:

∞
∞
(αk + 1) αk−α
D α yapp = Ck D α t αk = Ck t
(αk + 1 − α)
k=0 k=1
∞
(αk + α + 1) αk
= Ck+1 t ,
(αk + 1)
k=0
∞ ∞ ∞
2
1 (αk + α + 1) αk tk
I= Ck+1 t − Ck t αk − Bk (dt)α → min,
0 (αk + 1) k!
k=0 k=0 k=0
We apply now the minimization conditions
∂I
= 0 ⇒,
∂Ck
from which we obtain the relations:

1 ∞ ∞ ∞

(αk + α + 1) t k
t αk Ck+1 t αk − Ck t αk − Bk (dt)α = 0,
0 (αk + 1) k!
k=0 k=0 k=0
where k = 0, 1, . . .
∞
1 (αk + α + 1) 2αk Bk k+αk

Ck+1 t − Ck t 2αk
− t (dt)α = 0, k = 0, 1, . . .
(αk + 1) k!
k=0 0
We obtain from these relations:
C0 = B0 = 1,
(αk + α + 1) (2αk + 1) (2αk + 1)

Ck+1 +Ck
(αk + 1) (2αk + α + 1) (2αk + α + 1)
1 (αk + k + 1)
−Bk = 0,
k! (αk + k + 1 + α)
www.pdfgrip.com
1 (α + 2) (α + 1)(3α + 1)

C1 = 0 C2 = − ,
2 (2α + 2) 2 (2α + 1)
...
Finally, we obtain the solution:
1 (α + 2) (α + 1)(3α + 1) 2α

yapp (t) = 1 − t + ...
2 (2α + 2) 2 (2α + 1)
1
For α = 1, we have yapp ≈ 1 − t 2 .
4
Using the Maple sequence of commands:
MAPLE
> ec:=diff(y(t),t,t) + y(t) = t/(exp(t) -1 );
> dsolve({ec,y(0) = 1,D(y)0) = 0},y(t),series);
1
it results yapp ≈ 1 − t 3 .
12
Remark Due the fact that FI is difficult to calculate it is recommended that the
function f (t) to be represented by a series of powers. So:
∞ n
t
f (t) = et = ;
n!
n=0
, - ∞
cosh(t) (±1)n n
f (t) = = t
cos(t) (2n)!
n=0
, - ∞
sinh(t) (±1)n n
f (t) = = t
sin(t) (2n + 1)!
n=0
Example 3 Establish the solution, with the aid of least squares method, for the FDE:
4t 1/2
D (3/2) y(t) − t 3/2 y(t) = √ − t 7/2 ,
π
y(0) = 0, y (0) = 0.
Solution Let B = {B0 , B1 , B2 } be basis, with:

B0 = 1,
B1 = t,
B2 = t 2 .
www.pdfgrip.com
We consider a solution of type:
yap = xB0 + yB1 + zB2
Using the least squares method we can build the functional I , which will be
minimized:
1
1
4t 1/2
2
I= D yapp − t
α 3/2
yapp − √ − t 3/2 (dt)3/2 → min,
(5/2) 0 π
or:
1
1
I = (D 3/2 B0 − t 3/2 B0 ) x + (D 3/2 B1 − t 3/2 B1 ) y
(5/2) 0
2
4t 1/2
+(D 3/2 B2 − t 3/2 B2 ) z − √ − t 3/2 (dt)3/2 → min.
π
We use the notations

A = D 3/2 B0 − t 3/2 B0 ;
B = D 3/2 B1 − t 3/2 B1 ;
C = D 3/2 B2 − t 3/2 B2 .
The functional I becomes
1
1
4t 1/2
2
I= Ax + By + Cz − √ − t 3/2 (dt)3/2 → min,
(5/2) 0 π
and the minimization conditions are:

∂I ∂I ∂I
= 0, = 0, = 0.
∂x ∂y ∂z
Explicitly, we have:
⎧ 1 1/2
⎪
⎪
1 4t
⎪
⎪ [Ax + By + Cz] A (dt)3/2
= √ − t 3/2
A(dt)3/2 ,
⎪
⎪ π
⎪
⎪
0 0
⎪
⎪
⎪
⎨ 1 1 1/2
4t
[Ax + By + Cz] B (dt) =
3/2
√ −t 3/2
B (dt)3/2 ,
⎪
⎪ π
⎪
⎪
0 0
⎪
⎪
⎪
⎪ 1 1 1/2
⎪
⎪ 4t
⎪
⎩ [Ax + By + Cz] C (dt) 3/2
= √ − t 3/2
C(dt)3/2 .
0 0 π
www.pdfgrip.com
We obtain the system

⎧
⎪
⎪ A11 x + A12 y + A13 z = F1 ,
⎪
⎪
⎪
⎨
A21 x + A22 y + A23 z = F2 ,
⎪
⎪
⎪
⎪
⎪
⎩A x +A y +A z = F .
31 32 33 3
with:
1 1 1
A11 = A2 (dt)3/2 , A12 = B A(dt)3/2 , A13 = C A (dt)3/2 ,
0 0 0
1 4t 1/2
F1 = √ −t 3/2
A (dt)3/2
0 π
1 1 1
A21 = A B (dt)3/2 , A22 = B 2 (dt)3/2 , A23 = C B (dt)3/2 ,
0 0 0
1 4t 1/2
F2 = √ −t 3/2
B (dt)3/2
0 π
1 1 1
A31 = C A (dt)3/2 , A32 = C B (dt)3/2 , A13 = C 2 (dt)3/2 ,
0 0 0
1 4t 1/2
F3 = √ −t 3/2
C (dt)3/2 .
0 π
We observe that:
A12 = A21 , A13 = A31 , A23 = A32 ,

1 1 (β + 1)
f (β) = t β (dt)3/2 = .
(5/2) 0 (β + 5/2)
In MAPLE, we obtain exact solution y(t) = t 2 :
> restart;
> Digits:= 6;
> B0:= 1;
> B1:= t;
> B2:=t^2;
> A:= fracdiff(B0,t,3/2) - t^(3/2)*B0;
www.pdfgrip.com
> B:= fracdiff(B1,t,3/2) - t^(3/2)*B1;

> C:= fracdiff(B2,t,3/2) - t^(3/2)*B2;
> f:= proc(beta);
> f(beta):=evalf(GAMMA(beta+1)/GAMMA(beta+5/2))
> end proc;
> a11:= collect(A^2,t);
> A11:= f(3);
> a12:= collect(B*A,t);
> A12:= f(4);
> a13:= collect(C*A,t);
> A13:= f(5) - evalf(4/Pi^(1/2))*f(2);
> A21:= A12;
> a22:= collect(B^2,t);
> A22:= f(5);
> a23:= collect(C*B,t);
> A23:= f(6) - evalf(4/sqrt(Pi))*f(3);
> A31:= A13;
> A32:= A23;
> a33:= collect(C^2,t);
> A33:= f(7) - evalf(8/sqrt(Pi))*f(4) + evalf
(16/Pi)*f(1);
> f1:= collect((4/sqrt(Pi)*t^(1/2) - t^(7/2))*A,t);
> F1:= f(5) - evalf(4/sqrt(Pi))*f(2);
> f2:= collect(((4/sqrt(Pi))*t^(1/2) - t^(7/2))*B,t);
> F2:= f(6) - evalf(4/sqrt(Pi))*f(3);
> f3:= collect((4/sqrt(Pi)*t^(1/2) - t^(7/2))*C,t);
> F3:= f(7) - evalf(8/sqrt(Pi))*f(4)+11/Pi*f(1);
> ec1:= A11*x + A12*y + A13*z = F1;
> ec2:= A21*x + A22*y + A23*z = F2;
> ec3:= A31*x + A32*y + A33*z = F3;
> solve({ec1,ec2,ec3},{x,y,z});
The solution is {x = 0, y = 0, z = 1}, the exact solution being y(t) = t 2 . In

Mathematica we obtain:
MATHEMATICA
Clear["Global‘*"]
f[x_] := Gamma[x + 1]/Gamma[x + 5/2] // N
B0 = 1
B1 = t
B2 = t^2
A = -t^(3/2)*B0
B = -t^(3/2)*B1
C1 = 4*Sqrt[t] Sqrt[Pi] - t^(3/2)*B2
a11 = A^2
www.pdfgrip.com
6.3 The Galerkin Method for Fractional Differential Equations 135
Expand[%]
A11 = f[3]
a12 = B*A
Expand[%]
A12 = f[4]
a13 = C1*A
Expand[%]
A13 = f[5] - 4/Pi^(1/2)*f[2] // N
A21 = A12
a22 = B^2
Expand[%]
A22 = f[5]
a23 = C*B
Expand[%]
A23 = f[6] - 4/Sqrt[Pi]*f[3] // N
A31 = A13
A32 = A23
a33 = C1^2
Expand[%]
A33 = f[7] - 8/Sqrt[Pi]*f[4] + 16/Pi*f[1] // N
f1 = (4/Sqrt[Pi]*t^(1/2) - t^(7/2))*A
Expand[%]
F1 = f[5] - 4/Sqrt[Pi]*f[2] // N;
f2 = (4/Sqrt[Pi]*t^(1/2) - t^(7/2))*B
Expand[%]
F2 = f[6] - 4/Sqrt[Pi]*f[3] // N
f3 := (4/Sqrt[Pi]*t^(1/2) - t^(7/2))*C
F3 := f[7] - 8/Sqrt[Pi]*f[4] // N
Expand[%]
e3 = f[7] - 8/Sqrt[Pi]*f[4] // N
ec1 = A11*x + A12*y + A13*z
ec2 = A21*x + A22*y + A23*z
ec3 = A31*x + A32*y + A33*z
Solve[{ec1, ec2, ec3} == {F1, F2, F3}, {x, y, z}]
6.3 The Galerkin Method for Fractional Differential

Equations
The Galerkin3 method is a direct method for approximate estimation of the solution
of FDE. This method will be used here to the FDE:
3 B.G. Galerkin (1871–1945).

www.pdfgrip.com
D α y(t) + f (t) = 0,
where 0 < t < 1, and 0 < α ≤ 1, and the conditions:
y(0) = 0, y(1) = 0.
We denote by R(t) the residual of the equation
R(t) = D α yapp (t) + f (t),
and we will use the approximate solution

N
yapp (t) = Ci φi (t),
i=1
where φi (t) are the test (or weight) functions, so that

1
R(t)φi (t)(dt)α = 0, i = 1, 2, . . . , N
0
1
[D α yapp (t) + f (t)]φi (t)(dt)α = 0, i = 1, 2, . . . , N,
0
1 1
D α yapp (t)φi (t)(dt)α = − f (t)φi (t)(dt)α , i = 1, 2, . . . , N.
0 0
Example 1 Use the Galerkin method to find the approximate solution for the FDE:
D α y(t) − t = 0,
y(0) = 0, y (α) (0) = 0, 0 < α ≤ 1.
Solution We choose the test functions:
φ1 = t, φ2 = t 2 ,
resulting the solution:
yapp = C1 t + C2 t 2 .
www.pdfgrip.com
We obtain
D α yapp = C1 D α t + C2 D α t 2 .
It results the formulas:

(k + 1) k−α
Dα t k = t ,
(k + 1 − α)
(k + 1) k+α
I αtk = t ,
(k + 1 + α)
(1 + 1) 1−α (2 + 1) 2−α

D α yapp = C1 t + C2 t ,
(1 + 1 − α) (2 + 1 − α)
and finally, we obtain the system

⎧ 1 1
⎪
⎪
2 1−α
+
2 2−α α
= − t 2 (dt)α ,
⎪
⎪ C 1 t C 2 t t (dt)
⎨ 0 (2 − α) (3 − α) 0
⎪
⎪
⎪
1 2 2
1
⎪
⎩ C1 t 1−α
+ C2 t 2−α
t (dt) = −
2 α
t 3 (dt)α ,
0 (2 − α) (3 − α) 0
or:
⎧ 2 (3 − α) 2 (4 − α) (3)
⎪
⎪ C1 + C2 =− ,
⎪
⎨ (2 − α) (3) (3 − α) (4) (3 + α)
⎪
⎪ (4 − α) (5 − α)
⎪
⎩ C1 2 2 (4)
+ C2 =− .
(2 − α) (4) (3 − α) (5) (4 + α)
For:
α = 0.50, ⇒ C1 = −0.4156, C2 = 0.02536.
α = 1, ⇒ C1 = −0.3871, C2 = 0.08067.
Example 2 Establish an approximate solution for the following FDE, using the
Galerkin method:
D 2α y(t) + t = 0,
with the boundary conditions
y(0) = 0, y(1) = 0, 0 < α ≤ 1.
Solution We choose the test functions
v1 = t 2 (1 − t), v2 = t 2 (1 − t 2 ),
www.pdfgrip.com
and the approximate solution:
yapp = C1 v1 + C2 v2 .
We obtain
D 2α yapp = C1 D 2α v1 + C2 D 2α v2 ,
(k + 1)
D 2α t k = t k−2α ,
(k + 1 − 2α)
and the Galerkin system:

⎧ 1
⎪
⎪
⎪
⎪ D 2α yapp v1 (dt)α = 0,
⎨ 0
⎪
⎪
⎪
⎪
1
⎩ D 2α yapp v2 (dt)α = 0,
0
⎧ 1 1 1
⎪
⎪
⎪
⎪ C 1 D 2α
v1 v1 (dt) α
+ C 2 D 2α
v2 v1 (dt) α
= − tv1 (dt)α ,
⎨ 0 0 0
⎪
⎪
⎪
⎪
1 1 1
⎩ C1 D 2α v1 v2 (dt)α + C2 D 2α v2 v2 (dt)α = − tv2 (dt)α .
0 0 0
Introducing the notations
2 6 24
A= , B= , C= ,
(3 − 2α) (4 − 2α) (5 − 2α)
we have:
D 2α v1 = D 2α t 2 (1 − t) = At 2−2α − Bt 3−2α ,
D 2α v2 = D 2α (t 2 − t 4 ) = At 2−2α − Ct 4−2α ,
(D 2α v1 )v1 = At 4−2α − Bt 5−2α − At 5−2α + Bt 6−2α ,
(D 2α v1 )v2 = At 4−2α − Bt 5−2α − At 6−2α + Bt 7−2α ,
(D 2α v2 )v1 = At 4−2α − At 5−2α − Ct 6−2α + Ct 7−2α ,
(D 2α v2 )v2 = At 4−2α − At 6−2α − Ct 6−2α + Ct 8−2α ,

www.pdfgrip.com
1
B11 = (D 2α v1 )v1 (dt)α
0
(5 − 2α) (6 − 2α) (6 − 2α) (7 − 2α)
=A −B −A +B ,
(5 − α) (6 − α) (6 − α) (7 − α)
1
B12 = (D 2α v1 )v2 (dt)α
0
(5 − 2α) (6 − 2α) (7 − 2α) (8 − 2α)
=A −B −A +B ,
(5 − α) (6 − α) (7 − α) (8 − α)
1
B21 = (D 2α v1 )v2 (dt)α =
0
(5 − 2α) (6 − 2α) (7 − 2α) (8 − 2α)

=A −A −C +C ,
(5 − α) (6 − α) (7 − α) (8 − α)
1
B22 = (D 2α v2 )v2 (dt)α =
0
(5 − 2α) (7 − 2α) (7 − 2α) (9 − 2α)

=A −A −C +C ,
(5 − α) (7 − α) (7 − α) (9 − α)
1 (4) (5)
F1 = − tv1 (dt)α = − + ,
0 (4 + α) (5 + α)
1 (4) (6)
F2 = − tv2 (dt)α = − + .
0 (4 + α) (6 + α)
Using the notation

⎛ ⎞
B11 B12 ,
B=⎝ ⎠,
B21 B22 ,
C = [C1 C2 ],
F = [F1 F2 ],
we obtain in the matrix notation:
BC T = F T ⇒ C T = B −1 F T .
www.pdfgrip.com
Using Maple or Mathematica C1 and C2 can be calculated:

MAPLE
restart;Digits:=4:
> a:=1/2:
> V1:= t^2 - t^3:
> V2:= t^2 - t^4;
> A:= evalf(2/GAMMA(3 - 2*a));
> B:= evalf(6/GAMMA(4 - 2*a));
> C:= evalf(24/GAMMA(5 - 2*a));
> B11:=evalf(A*GAMMA(5 - 2*a)/GAMMA(5 - a)
> - B*GAMMA(6 - 2*a)/GAMMA(6 - a)
> - A*GAMMA(6 - 2*a)/GAMMA(6 - a)
> + B*GAMMA(7 - 2*a)/GAMMA(7 - a));
> - B*GAMMA(6 - 2*a)/GAMMA(6 - a)
> - A*GAMMA(7 - 2*a)/GAMMA(7 - a)
> + B*GAMMA(8 - 2*a)/GAMMA(8 - a));
> - A*GAMMA(6 - 2*a)/GAMMA(6 - a)
> - C*GAMMA(7 - 2*a)/GAMMA(7 - a)
> + C*GAMMA(8 - 2*a)/GAMMA(8 - a));
> - A*GAMMA(7 - 2*a)/GAMMA(7 - a)
> - C*GAMMA(7 - 2*a)/GAMMA(7 - a)
> + C*GAMMA(9 - 2*a)/GAMMA(9 - a));
> F1:=evalf(-GAMMA(4)/GAMMA(4 + a)
+ GAMMA(5)/GAMMA(5 + a));
> F2:=evalf(-GAMMA(4)/GAMMA(4 + a)
+ GAMMA(6)/GAMMA(6 + a));
> ec1:= C1*B11 + C2*B12 = F1;
> ec2:= C1*B21 + C2*B22 = F2;
> solve({ec1,ec2},{C1,C2});
MATHEMATICA
a= 1/2
A = 2/Gamma[3 - 2*a]//N
B = 2/Gamma[4 - 2*a]//N
C = 24/Gamma[5 - 2*a]//N
B11 = A*Gamma[5 - 2*a]/Gamma[5 - a]
- B*Gamma[6 - 2*a]/Gamma[6 - a]
- A*Gamma[6 - 2*a]/Gamma[6 - a]
+ B*Gamma[7 - 2*a]/Gamma[7 - a]//N
- B*Gamma[6 - 2*a]/Gamma[6 - a]
www.pdfgrip.com
+ B*Gamma[8 - 2*a]/Gamma[8 - a]//N
- C*Gamma[7 - 2*a]/Gamma[7 - a]
+ C*Gamma[8 - 2*a]/Gamma[8 - a]//N
- C*Gamma[7 - 2*a]/Gamma[7 - a]
+ C*Gamma[9 - 2*a]/Gamma[9 - a]//N
F1 = -Gamma[4]/Gamma[4 + a] + Gamma[5]/
Gamma[5 + a]//N
F2 = -Gamma[4]/Gamma[4 + a] + Gamma[6]/
Gamma[6 + a]//N
ec1 = C1*B11 + C2*B12
ec2 = C1*B12 + C2*B22
Solve[{ec1,ec2}=={F1,F2},{F1,F2}]//N
For:
α = 0.50, C1 = −12.89, C2 = 8.135
α = 1, C1 = 1.067, C2 = −0.3944
Example 3 Establish an approximate solution using the Galerkin method, for the
FDE:
D 1/2 y(t) = −1 + y 2 (t), y(0) = 0, 0 < t ≤ 1.
Solution We apply the LT and Adomian method :
L[D α y(t)] = −L[1] + L[y 2 (t)], y(0) = 0,
(see Example 2 presented in Sec. 4.3.2) solution is:
tα (2α + 1)
Ya = − + 2 t 3α + . . .
(α + 1) (α + 1)(3α + 1)
We consider approximate solution
Ya = At 1/2 + Bt 3/2 ,
and we have:
⎧ 1 1
⎪
⎪
⎪
⎪ (D 1/2
Y a − Y 2 1/2
)t (dt) 1/2
= − t 1/2 (dt)1/2 ,
⎨ 0 a
0
⎪
⎪ 1
⎪
⎪
1
⎩ (D Ya − Ya )t (dt) = −
1/2 2 3/2 1/2
t 3/2 (dt)1/2 .
0 0
www.pdfgrip.com
Using Maple or Mathematica A and B can be calculated:

MAPLE
> restart:
> Digits:=4:
> a:=1/2:
> Y_a:=A*tâ+B*t^(3*a):
> fracdiff(Y_a,t,1/2):
> unassign(’t’):
> g:=t->fracdiff(Y_a,t,1/2)-(Y_a)^2:
> ex1:=g(t)*t^(1/2):
> Ex1:=collect(ex1,t):
> ex2:=g(t)*t^(3/2):
> Ex2:=collect(ex2,t):
> f:=t->GAMMA(t+1)/GAMMA(t+1+1/2):
> f1:=f(1/2):
> f2:=f(3/2):
> ec1:=-B^2*f(7/2)-2*A*B*f(5/2)+(3/4*Pi^(1/2)
*B-A^2)*f(3/2)+1/2*Pi^(1/2)*A*f(1/2)=-f1:
> ec2:=-B^2*f(9/2)-2*A*B*f(7/2)+(3/4*Pi^(1/2)
*B-A^2)*f(5/2)+1/2*Pi^(1/2)*A*f(3/2)=-f2:
> ecu1:=evalf(ec1):
> ecu2:=evalf(ec2):
> solve({ecu1,ecu2},{A,B});
{A = -0.9740, B = 0.2786}, {A = -5.798, B = 9.315},
{A = 4.568 - 0.6385 I, B = -3.956 + 1.690 I},
{A = 4.568 + 0.6385 I, B = -3.956 - 1.690 I}
MATHEMATICA
Clear["‘*"]
f[x_] := Gamma[x + 1]/Gamma[x + 3/2];
Ya = A*t^(1/2) + B*t^(3/2);
DYa = A*Sqrt[Pi]/2 + B*3*Sqrt[Pi]/4*t;
g[t_] := DYa - (Ya)^2;
ex1 = g[t]*t^(1/2);
Ex1 = Expand[%];
ex2 = g[t]*t^(3/2);
Ex2 = Expand[%];
f1 = -f[1/2] // N;
f2 = -f[3/2] // N;
ex1 = (Sqrt[Pi]/2*A + 3*Sqrt[Pi]/4*B - (A*Sqrt[t]+
B*t^(3/2))^2)*Sqrt[t];
Ex1 = Expand[%];
ex2 = (Sqrt[Pi]/2*A + 3*Sqrt[Pi]/4*B - (A*Sqrt[t]+
B*t^(3/2))^2)*t^(3/2);
www.pdfgrip.com
6.4 Euler’s Method 143
Ex22 = Expand[%];
ec1 = 1/2*A*Sqrt[Pi]*f[1/2] + 3/4*B*Sqrt[Pi]*f[3/2]
- A^2*f[3/2] - 2 A*B*f[5/2] - B^2*f[7/2] // N;
ec2 = 1/2*A*Sqrt[Pi]*f[3/2] + 3/4*B*Sqrt[Pi]*f[5/2]
- A^2*f[5/2] - 2 A*B*f[7/2] - B^2*f[9/2] // N;
NSolve[{ec1, ec2} == {f1, f2}, {A, B}]
6.4 Euler’s Method
We introduce here a generalization of the Euler method to the case of FDE of type:
D α y(t) = f (t, y(t)), with: y(t0 ) = y0 , where: 0 < α ≤ 1.
We assume that y(t), D α y(t), D 2α y(t) are continuous on [t0 , a] and we will
apply the generalized Taylor’s formula:
tα t 2α
y(t) = y(t0 ) + D α y(t0 ) + D 2α (η) , t0 < η ≤ t.
(α + 1) (2α + 1)
We obtain the iterative Euler’s formula:
hα
y(t) ≈ y(t0 ) + f (t0 , y(t0 )),
(α + 1)
or, expressed in the recurrent form:
hα
yn+1 ≈ yn + f (tn , yn ),
(α + 1)
Example 1 Solve the following FDE:
2t
D 1/2 y(t) = y(t) − , with the initial condition: y(0) = 1,
y(t)
where t ∈ [0, 1], and step h = 0.2.

Solution We can calculate the solution in Maple and Mathematica using the Euler
formula:
MAPLE
> restart;
> Digits:=4:
> unassign(t,y);
> f:=(t,y)->y-2*t/y:
www.pdfgrip.com
> d:=0.2:
> a:=1/2:
> h:=evalf(dâ/GAMMA(a+1)):
> t:=array[0..10]:
> y:=array[0..10]:
> for k from 0 to 5 do t[k]:=k*0.2 od:
> y[0]:=1:
> for k from 0 to 5 do y[k+1]:=y[k]+h*f(t[k],y[k]) od:
> for k from 0 to 5 do print(t[k]," " ,y[k]) od;
0.0, 1
0.2, 1.504
0.4, 2.128
0.6, 3.012
0.8, 4.331
1.0, 6.329
MATHEMATICA
Clear["*"]
\[Delta] = 0.1; \[Alpha] = 1/2;
f[t_, y_] := y - 2*t/y
values =
RecurrenceTable[{t[k + 1] == t[k] + \[Delta],
y[k + 1] ==
y[k] + \[Delta]^\[Alpha]/Gamma[\[Alpha] + 1]
*f[t[k], y[k]],
t[0] == 0, y[0] == 1}, {t, y}, {k, 0, 10}];
Grid[values]
ListLinePlot[values, PlotMarkers -> Automatic]
The solution y(t) is plotted in Fig. 6.1.
Example 2 Approximate the following FDE with the aid of Euler method:
y(t) − t
D α y(t) = , for the initial condition: y(0) = 1,
y(t) + t
with the step h = 0.1, and t ∈ [0, 1].

MAPLE
< restart;
> Digits:=4:
> unassign(t,y);
> f:=(t,y)->(y- t)/(y+t):
> d:=0.1:
> a:=1/2:
www.pdfgrip.com
Fig. 6.1 The solution y(t) of

Example 1
15
10
0.2 0.4 0.6 0.8 1.0
> t:=array[0..10]:
> y:=array[0..10]:
> for k from 0 to 10 do t[k]:=k*0.1 od:
> y[0]:=1:
> for k from 0 to 10 do y[k+1]:=y[k]+h*f(t[k],
y[k]) od:
> for k from 0 to 10 do print(t[k]," " ,y[k]) od;
0., 1
0.1, 1.357
0.2, 1.665
0.3, 1.945
0.4, 2.206
0.5, 2.453
0.6, 2.689
0.7, 2.916
0.8, 3.135
0.9, 3.347
1.0, 3.552
MATHEMATICA
Clear["*"]
\[Delta] = 0.1; \[Alpha] = 1/2;
f[t_, y_] := (y - t)/(y + t)
values =
RecurrenceTable[{t[k + 1] == t[k] + \[Delta],
www.pdfgrip.com
y[k + 1] ==
y[k] + \[Delta]^\[Alpha]/Gamma[\[Alpha] + 1]
*f[t[k], y[k]],
t[0] == 0, y[0] == 1}, {t, y}, {k, 0, 10}];
Grid[values]
ListLinePlot[values, PlotMarkers -> Automatic]
0.0 1
0.1 1.35582
0.2 1.86466
0.3 1.94494
0.4 2.2064
0.5 2.4532
0.6 2.68972
0.7 2.91639
0.8 3.13587
0.9 3.34681
1 3.5524
In this case the solution y(t) is plotted in Fig. 6.2.
Using the eulerstep[. . . ] command, the solution can be obtained using the
Mathematica program:
MATHEMATICA
Clear["‘*"]
a = 1/2;
d = 0.1;
eulerstep[f_, {t_, y_}, h_] := {t + h,
Fig. 6.2 The solution y(t) of

3.5
Example 2
3.0
2.5
2.0
1.5
0.2 0.4 0.6 0.8 1.0

www.pdfgrip.com
y + d â/Gamma[a + 1] f[t, y]}

euler[f_, {t_, y_}, tf_, h_] :=
NestList[eulerstep[f, #1, h] &, {t, y},
Ceiling[(tf - t)/h]]
f[t_, y_] := (y - t)/(y + t)
tmp = euler[f, {0, 1}, 1, 0.1];
PaddedForm[TableForm[tmp], {6, 4}]
0.0, 1.0,
0.1, 1.3568,
0.2, 1.6646,
0.3, 1.9449,
0.4, 2.2064,
0.5, 2.4537,
0.6, 2.6897,
0.7, 2.9163,
Example 3 Approximate the system of FDE:
⎧
⎪ y(t) − t
⎪
⎨ D x(t) =
1/2
⎪ ,
t
⎪
⎪
⎩ D 1/2 y(t) = x(t) + t ,
⎪
t
with the initial conditions x(1) = 1, y(1) = 1, the time step h = 0.2 and t ∈ [1, 2].
Solution We can use the following program:

MAPLE
> restart; Digits:=4:
> unassign(t,x,y);
> f:=(t,x,y)->(y-x)/t:g:=(t,x,y)->(y+x)/t:
> d:=0.2:
> a:=1/2:
> t:=array[1..10]:x:=array[1..10]:
> y:=array[1..10]:t[1]:=1:
> for k from 1 to 6 do t[k+1]:=t[k]+0.2 od:
> y[1]:=1:x[1]:=1:
x[k+1]:=x[k]+h*f(t[k],x[k],y[k]):
y[k+1]:=y[k]+h*g(t[k],x[k],y[k]) od;
> for k from 1 to 6 do print(t[k],x[k] ,y[k]) od;
1,0 1, 1
1.2, 1., 2.009
1.4, 1.424, 3.274
1.6, 2.090, 4.967
www.pdfgrip.com
1.8, 2.997, 7.192

2.0, 4.173, 10.05
MATHEMATICA
Clear["*"]
\[Delta] = 0.2; \[Alpha] = 1/2;
f[t_, x_, y_] := (y - x)/t
g[t_, x_, y_] := (y + x)/t
values = RecurrenceTable[{t[k + 1] == t[k] + \[Delta],
x[k + 1] == x[k] + \[Delta]^\[Alpha]/Gamma[\[Alpha]
+ 1]*f[t[k], x[k], y[k]],
y[k + 1] == y[k] + \[Delta]^\[Alpha]/Gamma[\[Alpha]
+ 1]*g[t[k], x[k], y[k]],
t[0] == 1, x[0] == 1, y[0] == 1}, {t, x, y},
{k, 0, 5}]//
TableForm
1 1 1
1.2 1 2.00925
1.4 1.42441 3.27471
1.6 2.09135 4.96135
1,8 2.91878 7.19151
2 4.17521 10.053
Remark Using the notations
k1 = f (tn , yn ),

h k1
k2 = f tn + , yn + ,
2 2
a more general Euler’s iterative formula can be written as:
yn+1 = yn + h k2 .
Example 4 (Lorenz Attractor) Solve the Lorentz attractor system:

⎧
⎪
⎪ D 0.98 x(t) = −10(x − y), x(0) = 0,
⎪
⎪
⎪
⎨
D 0.98 y(t) = 28x − y − xz, y(0) = 1,
⎪
⎪
⎪
⎪
⎪
⎩ D 0.98 z(t) = xy − 8 z,
3 z(0) = 0.
The solution in Mathematica is:
Clear["‘*"]
d = 0.01
www.pdfgrip.com
6.5 Runge–Kutta Methods for Fractional Differential Equation 149
Fig. 6.3 The 3D Lorenz

attractor solution
(x(t), y(t), z(t)) for α = 0.98
a = 0.98
h = dâ/Gamma[a + 1]
Eulerlor[{x_, y_, z_}] := {x - 10*h*(x - y),
y + h*(28*x - y - x*z),
z + h*(x*y - 8/3*z)}
sol = NestList[Eulerlor, {0, 1, 0}, 1000];
ParametricPlot3D[Through@p@t, {t, 1, 1000},
PlotPoints -> 100,
ColorFunction -> (Hue[4 #] &), ImageSize -> 300]
Figure 6.3 shows the 3D Lorenz attractor solution (x(t), y(t), z(t)) for α = 0.98.
6.5 Runge–Kutta Methods for Fractional Differential

Equation
In this section we will suppose that all conditions of existence and uniqueness of the
solutions are fulfilled.
www.pdfgrip.com
6.5.1 The Second Order Runge–Kutta Method
The second order Runge4 –Kutta5 method Second order Runge–Kutta (RK2)
introduced here is an extension of the first Euler method for approximation of the
solution of the FDE:
D α y(t) = f (t, y(t)), y(t0 ) = y0 , 0 < α ≤ 1,
where:
y ∈ C p+1 ([t0 , t0 + T ]).
In this case the solution can be approximated in the following discrete form:
hα K1 + K2
yn+1 = yn + ,
(α + 1) 2
where
K1 = f (tn , yn ),

hα hα
K2 = f tn + , yn + K1 .
(α + 1 (α + 1)
Proof We introduce the notations
K1 = f (tn , yn ),

hα hα
K2 = f tn + A , yn + B K1 ,
(α + 1 (α + 1
where A and B are two unknown real constants.

The local error from the Taylor expansion is
E(h) = yn+1 − yn ,
with:
yn+1 = y(tn + h), tn+1 = tn + h, yn = y(tn ).
4 C.D.T. Runge (1856–1927).

5 M.W. Kutta (1867–1944).
www.pdfgrip.com
Obviously, for minimization of error, we impose the conditions:
E(0) = D α E(0) = . . . = D pα E(0) = 0, D (p+1)α = 0.
Using the Taylor polynomial we obtain:
hα h2α
E(h) = yn+1 − yn = D α yn + D 2α yn + O(h3α )
(α + 1) (2α + 1)
hα h2α
= f (tn , yn ) + D 2α yn + O(h3α ),
(α + 1) (2α + 1)
hα
D α E(h) = f (tn , yn ) + D α yn + O(h2α ),
(α + 1)
and
hα hα
yn+1 − yn = C1 K1 + C2 K2 .
(α + 1) (α + 1)
Hence, it results E[h] and D α E[h]:
hα hα
E[h] = C1 K1 + C2 K2 ,
(α + 1) (α + 1)
hα
D α E[h] = c1 K1 + c2 K2 + D α K2 ,
(α + 1)

hα hα
D α K2 = D α f tn + A , yn + B K1
(α + 1 (α + 1
= ftn +A hα A + fyn +B hα BK1 .

(α+1) (α+1) K1
For h → 0 we obtain:
D α E[0] = C1 K1 [0] + C2 K2 [0],
f (tn , yn ) = C1 f (tn , yn ) + C2 f (tn , yn ), ⇒ 1 = C1 + C2 ,
D 2α E(h) = D α (D α E(h)) = C2 D α K2 + D α K2 .
But D 2α E(0) = D 2α yn (t), and:
D 2α y(t) = D α (f (t, y(t)) = ft + fy D α y(t) = ft + ffy .

www.pdfgrip.com
It results:
2C2 Aftn (tn , yn ) + 2C2 Bf (tn , yn )fyn (tn , yn ) = ftn (tn , yn ) + f (tn , yn )fyn (tn , yn ).
We obtain finally the system:

⎧
⎪
⎪ C1 + C2 = 1
⎪
⎪
⎪
⎨
2C2 A − 1 = 0
⎪
⎪
⎪
⎪
⎪
⎩ 2C B − 1 = 0
2
1 hα
For C1 = C2 = we have the classical RK2 method, with the step .
2 (α + 1)
Example Solve with RK2 the FDE
y2
D 1/2 y(t) = t 2 + , y(0) = 1.
4
The solution can be written in Maple or Mathematica:
MAPLE
> restart;
> Digits:=3:
> a:=1/2:
> d:=0.2:
> y:=Array[0..10]:
> y[0]:=1:
> unassign(’t,y’);
> f:=(t,y)->t^2+y^2/4:
> for i from 0 to 4 do
> k1:=f(i*0.2,y[i]):
> k2:=f(i*0.2 + h,y[i]+h*k1):
> y[i+1]:=y[i]+(k1+k2)/2 od:
> for i from 0 to 5 do print(i*0.2,y[i]) od;
0 1
0.2 1.207
0.4 1.560
0.6 2.200
0.8 3.468
1 6.597
www.pdfgrip.com
MATHEMATICA
Clear["‘*"]
a = 1/2
d = 0.2
h = dâ/Gamma[a + 1]
y[0] = 1
f[t_, y_] := t^2 + y^2/4
For[i = 0, i < 5,
i++, {k1 = f[t, y] /. {t -> i*0.2, y -> y[i]},
k2 = f[t, y] /. {t -> i*0.2 + h, y -> y[i] + h*k1},
y[i + 1] = y[i] + (k1 + k2)*0.5*h}]
Table[{j*0.2, y[j]}, {j, 0, 5}] // TableForm
0 1
0.2 1.20733
0.4 1.56029
0.6 2.20011
0.8 3.469
1 6.59765
6.5.2 The Fourth Order Runge–Kutta Method
In the case of fourth order Runge–Kutta Fourth order Runge–Kutta (RK4)

algorithm, we will establish an algorithm to solve the following FDE:
D α y(t) = f (t, y(t)), y(t0 ) = y0 , 0 < α ≤ 1,
where:
y ∈ C p+1 ([t0 , t0 + T ]).
In the neighborhood of t0 we suppose that
D α E(0) = D 2α E(0) = 0, E(h) = y(t + h) − y(t).
The approximate solution can be obtained from the expansion
hα
yn+1 = yn + (K1 + 2K2 + 2K3 + K4 ),
6(α + 1)
where K1 , . . . , K4 are functions which will be established in the proof.

Proof The proof is similar to the previous case (RK2).
www.pdfgrip.com
We introduce here the expansion error:
hα h2α
E(h) = yn+1 − yn = D α yn + D 2α yn + O(h3α )
(α + 1) (2α + 1)
hα h2α
= f (tn , yn ) + D 2α yn + O(h3α ),
(α + 1) (2α + 1)
and its derivative of order α:
hα
D α E(h) = f (tn , yn ) + D α yn + O(h2α ),
(α + 1)
and
hα hα hα hα
yn+1 −yn = C1 K1 +C2 K2 +C3 K3 +C4 K4 .
(α + 1) (α + 1) (α + 1) (α + 1)
Hence, we have
hα hα hα hα
E[h] = C1 K1 + C2 K2 + C3 K3 + C4 K4 ,
(α + 1) (α + 1) (α + 1) (α + 1)
where:
K1 = f (tn , yn ),

hα hα
K2 = f tn + a2 , yn + b2 K1 ,
2(α + 1) 2(α + 1)

hα hα
K3 = f tn + a3 , yn + b3 K2 ,
2(α + 1) 2(α + 1)

hα hα
K4 = f tn + a4 , yn + b4 K3 ,
(α + 1) (α + 1)
and finally:
D α E[h] = C1 K1 + C2 K2 + C3 K3 + C4 K4
hα hα
+C1 D α K1 + C 2 D α K2
(α + 1) (α + 1)
hα hα
+C3 D α K3 + C 4 D α K4 .
(α + 1) (α + 1)
www.pdfgrip.com
We take now h → 0:
D α E[0] = C1 K1 [0] + C2 K2 [0] + C3 K3 [0] + C4 K4 [0]
f (tn , yn ) = (C1 + C2 + C3 + C4 )f (tn , yn ).
Hence, we obtain the equation:
1 = C1 + C2 + C3 + C4 .
From the other conditions we get:

⎧
⎨ C2 a2 + C3 a3 + C4 a4 = 2 ,
1
⎩
C2 b2 + C3 b3 + C4 b4 = 12 ,
(i.e., from D α E(h) = 0).

Finally, we have the system:
⎧
⎪
⎪ C1 + C2 + C3 + C4 = 1,
⎪
⎪
⎪
⎨
C2 a2 + C3 a3 + C4 a4 = 12 ,
⎪
⎪
⎪
⎪
⎪
⎩
C2 b2 + C3 b3 + C4 b4 = 12 .
Remark The solution of this system is not unique. For this reason we choose as in
the RK2 case:
C1 = C4 = 1/6, C2 = C4 = 1/3
1
a1 = b1 = a4 = b4 = 1, a2 = b2 = a3 = b3 =
2
Example 1 Solve, using the RK4 algorithm, the following FDE:
y2
D α y(t) = t 2 + , y(0) = 1,
4
for the cases:
(a) α = a = 12 ,
(b) α = a = 0.98.
www.pdfgrip.com
The solution can be written as:

MATHEMATICA
Clear["‘*"];
a = 1/2;
y[0] = 1;
h = (0.01)â/Gamma[a + 1];
n = 5;
f[t_, y_] = t^2 + y^2/4;
Do[{K1 = f[i h, y[i]], K2 = f[i*h + h/2, y[i]
+ h/2*K1],
K3 = f[i h + h/2, y[i]+h*K2/2], K4 =f[i h + h,
y[i]+h*K3],
y[i + 1] = y[i] + (K1 + 2 K2 + 2 K3 + K4)*h/6},
{i, 0, n}]
Do[Print[PaddedForm[i h, {6, 4}], " ",
PaddedForm[y[i], {6, 4}]], {i, 0, 5}];
0.0000 1.0000
0.1128 1.0295
0.2257 1.0637
0.3385 1.1060
0.4514 1.1599
0.5642 1.2292
Clear["‘*"]
RK4step[{t_, y_}] := Module[{k1, k2, k3, k4},
k1 = f[t, y];
k2 = f[t + h/2, y + h/2 k1];
k3 = f[t + h/2, y + h/2 k2];
k4 = f[t + h, y + h k3];
{t + h, y + 1/6 (k1 + 2 k2 + 2 k3 + k4)}]
f[t_, y_] := t^2 + y^2/4;
t0 = 0;
y0 = 1;
a = 1/2;
h = (0.01)â/Gamma[a + 1];
n = 5;
rkpoints = NestList[RK4step, {t0, y0}, n];
PaddedForm[TableForm[rkpoints], {6, 4}]
0, 1,
0.1128, 1.2616,
0.2256, 1.7047,
0.3385, 2.5522,
0.4513, 4.4782,
0.5641, 10.500
www.pdfgrip.com
Example 2 Solve the following system of FDE:

⎧ 2
⎪
⎪ 1/2 x(t) = t 2 + y , x(0) = 0,
⎪
⎨ D
4
⎪
⎪
⎪ 2
⎩ D 1/2 y(t) = t 2 + x , y(0) = 1.
4
Solution
MATHEMATICA
Clear["‘*"];
a = 1/2; x[0] := 0;
y[0] = 1; h = (0.01)â/Gamma[a + 1]; n = 5;
Do[t[i] = 0.0 + (0.2)*i, {i, 0, n}]
f[t_, x_, y_] = t^2 + y^2/4
g[t_, x_, y_] = t^2 + x^2/4
Do[{K1 = f[t[i], x[i], y[i]],
L1 = g[t[i], x[i], y[i]],
K2 = f[t[i] + h, x[i] + h*K1, y[i] + h*L1],
L2 = g[t[i] + h, x[i] + h*K1, y[i] + h*L1],
K3 = f[t[i] + h/2, x[i] + h*K2/2, y[i] + h*L2/2],
L3 = g[t[i] + h/2, x[i] + h*K2/2, y[i] + h*L2/2],
K4 = f[t[i] + h, x[i] + h*K3, y[i] + h*L3],
L4 = g[t[i] + h, x[i] + h*K3, y[i] + h*L3],
x[i + 1] = x[i] + (K1 + 2 K2 + 2 K3 + K4)*h/6,
y[i + 1] = y[i] + (L1 + 2 L2 + 2 L3 + L4)*h/6},
{i, 0, n}]
Do[Print[t[i], " ", x[i], " ", y[i]], {i, 0, 5}];
0 0 1
0.2 0.0290 0.0088
0.4 0.0663 1.0096
0.6 0.1216 1.0356
0.8 0.3054 1.0881
1 0.3189 1.1771
Example 3 Solve the fractional Van der Pol6 system:

⎧ 1/2
⎨ D x(t) = y, x(0) = 1,
⎩
D 1/2 y(t) = −x + 0.25 (1 − x 2 ) y, y(0) = 0.
6 B. van der Pol (1889–1959).

www.pdfgrip.com
Solution
MATHEMATICA
Clear["‘*"];
a = 1/2; x[0] := 1;
y[0] = 0; h = (0.2)â/Gamma[a + 1]; n = 10000;
Do[t[i] = 0.0 + (0.2)*i, {i, 0, n}]
f[t_, x_, y_] = y
g[t_, x_, y_] = -x + 0.25*(1 - x^2)*y
Do[{K1 = f[t[i], x[i], y[i]],
L1 = g[t[i], x[i], y[i]],
K2 = f[t[i] + h, x[i] + h*K1, y[i] + h*L1],
L2 = g[t[i] + h, x[i] + h*K1, y[i] + h*L1],
K3 = f[t[i] + h/2, x[i] + h*K2/2, y[i] + h*L2/2],
L3 = g[t[i] + h/2, x[i] + h*K2/2, y[i] + h*L2/2],
K4 = f[t[i] + h, x[i] + h*K3, y[i] + h*L3],
L4 = g[t[i] + h, x[i] + h*K3, y[i] + h*L3],
x[i + 1] = x[i] + (K1 + 2 K2 + 2 K3 + K4)*h/6,
y[i + 1] = y[i] + (L1 + 2 L2 + 2 L3 + L4)*h/6},
{i, 0, n}];
Do[Print[t[i], " ", x[i], " ", y[i]], {i, 0, 5}];
Figure 6.4 shows the plot of the Van der Pol solution from Example 3.
Fig. 6.4 Plot of the Van der

Pol solution from Example 3
1.0
0.5
0.0
–0.5
–1.0
–1.0 –0.5 0.0 0.5 1.0

www.pdfgrip.com
Example 4 Solve the fractional Duffing7 system:

⎧ 0.998
⎨D x(t) = y, x(0) = 1,
⎩
D 0.998 y(t) = −x − x 3 , y(0) = 0.
Solution
Clear["‘*"];
a = 0.998; x[0] = 1;
y[0] = 0; h = (0.2)â/Gamma[a + 1]; n = 10000;
Do[t[i] = 0.0 + (0.2)*i, {i, 0, n}]
f[t_, x_, y_] := y
g[t_, x_, y_] := -x - x^3
Do[{K1 = f[t[i], x[i], y[i]], L1 = g[t[i], x[i],
y[i]],
K2 = f[t[i] + h, x[i] + h*K1, y[i] + h*L1],
L2 = g[t[i] + h, x[i] + h*K1, y[i] + h*L1],
K3 = f[t[i] + h/2, x[i] + h*K2/2, y[i] + h*L2/2],
L3 = g[t[i] + h/2, x[i] + h*K2/2, y[i] + h*L2/2],
K4 = f[t[i] + h, x[i] + h*K3, y[i] + h*L3],
L4 = g[t[i] + h, x[i] + h*K3, y[i] + h*L3],
x[i + 1] = x[i] + (K1 + 2 K2 + 2 K3 + K4)*h/6,
y[i + 1] = y[i] + (L1 + 2 L2 + 2 L3 + L4)*h/6},
{i, 0, n}]; ListPlot[Table[{x[n], y[n]},
{n, 0, 10000}], Frame -> True]
Figure 6.5 shows the plot of the fractional Duffing solution from the Example 4.
Example 5 Solve fractional system
⎧
⎪
⎪ D 1/2 x(t) = 2y, x(0) = 1,
⎪
⎪
⎪
⎨
D 1/2 y(t) = 2z, y(0) = 1,
⎪
⎪
⎪
⎪
⎪
⎩ 1/2
D z(t) = x − y, z(0) = 1.
Solution
MAPLE
> restart;
> Digits:=4:X:=array[0..5]:Y:=array[0..5]:
Z:=array[0..5]:
7 Ge. Duffing (1861–1944).

www.pdfgrip.com
1.x10–10
5.x10–11
–5.x10–11
–1.x10–10
–1.x10–10 –5.x10–11 0 5.x10–11 1.x10–10
Fig. 6.5 Plot of the fractional Duffing solution from the Example 4
> unassign(’t,x,y,z’):
> f:=(t,x,y,z)->2*y:
> g:=(t,x,y,z)->2*z:
> p:=(t,x,y,z)->x - y:
> a:=1/2: x:=1:X[0]:=1: y:=1:Y[0]:=1:
> z:=1:Z[0]:=1: d:=0.2:
> for n from 0 to 5 by 1 do
> k1:=h*f(d*n,x,y,z);
> l1:=h*g(d*n,x,y,z):
> m1:=h*p(d*n,x,y,z):
> k2:=h*f(d*n+h,x+k1,y+l1,z+m1):
> l2:=h*g(d*n+h,x+k1,y+l1,z+m1):
> m2:=h*p(d*n+h,x+k1,y+l1,z+m1);
> k3:=h*f(d*n+h/2,x+k2/2,y+l2/2,z+m2/2):
> l3:=h*g(d*n+h/2,x+k2/2,y+l2/2,z+m2/2):
> m3:=h*p(d*n+h/2,x+k2/2,y+l2/2,z+m2/2):
> k4:=h*f(d*n,x+k3,y+l3,z+m3):
> l4:=h*g(d*n,x+h*k3,y+h*l3,z+h*m3):
www.pdfgrip.com
> m4:=h*p(d*n,x+h*k3,y+h*l3,z+h*m3):
> x:=x+(k1+2*k2+2*k3+k4)/6:X[n+1]:=x:
> y:=y+(l1+2*l2+2*l3+l4)/6:Y[n+1]:=y:
> z:=z+(m1+2*m2+2*m3+m4)/6:Z[n+1]:=z od:
> for n from 0 to 5 do print(d*n,X[n],Y[n],Z[n]) od;
0., 1, 1, 1
0.2, 2.688, 2.030, 1.108
0.4, 5.612, 3.458, 1.795
0.6, 10.67, 6.102, 3.458
0.8, 19.87, 11.26, 6.694
1.0, 37.08, 21.11, 12.70
MATHEMATICA
Clear["‘*"]
f[t_, x_, y_, z_] := 2*y;
g[t_, x_, y_, z_] := 2*z;
p[t_, x_, y_, z_] := x - y;
a = 1/2; d = 0.2; x[0] = 1; y[0] = 1; z[0] = 1;
tmax = 5;
h = (d)â/Gamma[a + 1];
Do[{k1 = h f[t[n], x[n], y[n], z[n]];
l1 = h g[n*d, x[n], y[n], z[n]];
m1 = h p[n*d, x[n], y[n], z[n]];
k2 = h f[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
l2 = h g[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
m2 = h p[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
k3 = h f[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
l3 = h g[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
m3 = h p[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
k4 = h f[n*d + h, x[n] + k3, y[n] + l3, z[n] + m3];
l4 = h g[n*d + h, x[n] + k3, y[n] + l3, z[n] + m3];
m4 = h p[n*d + h, x[n] + k3, y[n] + l3, z[n] + m3];
x[n + 1] = x[n] + 1/6 *(k1 + 2 k2 + 2 k3 + k4);
y[n + 1] = y[n] + 1/6* (l1 + 2 l2 + 2 l3 + l4);
z[n + 1] = z[n] + 1/6* (m1 + 2 m2 + 2 m3 + m4)};
Print[0.2*n, " ", x[n], " ", y[n], " ", z[n]],
{n, 0, tmax}]
www.pdfgrip.com
Example 6 (Lorenz Attractor with Interpolation Solution) Solve the Lorenz attrac-
tor problem:
⎧
⎪
⎪D
0.998 x(t) = 10(y − x), x(0) = 1,
⎪
⎪
⎪
⎨
D 0.998 y(t) = 28x − y − xz, y(0) = 1,
⎪
⎪
⎪
⎪
⎪
⎩ D 0.998 z(t) = xy − 8 z,
3 z(0) = 1,
in Mathematica, using the interpolation command.
Clear["‘*"]
f[t_, x_, y_, z_] := 10*(y - x);
g[t_, x_, y_, z_] := 28*x - y - x*z;
p[t_, x_, y_, z_] := x*y - 8/3* z;
a = 0.998; d = 0.01;
x[0] = 1;
y[0] = 1;
z[0] = 1;
tmax = 10000;
h = (d)â/Gamma[a + 1]; Do[{k1 = h f[t[n], x[n],
y[n], z[n]];
l1 = h g[n*d, x[n], y[n], z[n]];
m1 = h p[n*d, x[n], y[n], z[n]];
k2 = h f[n*d + h/2, x[n] + k1/2, y[n] + l1/2, z[n]
+ m1/2];
l2 = h g[n*d + h/2, x[n] + k1/2, y[n] + l1/2, z[n]
+ m1/2];
m2 = h p[n*d + h/2, x[n] + k1/2, y[n] + l1/2, z[n]
+ m1/2];
k3 = h f[n*d + h/2, x[n] + k2/2, y[n] + l2/2, z[n]
+ m2/2];
l3 = h g[n*d + h/2, x[n] + k2/2, y[n] + l2/2, z[n]
+ m2/2];
m3 = h p[n*d + h/2, x[n] + k2/2, y[n] + l2/2, z[n]
+ m2/2];
x[n + 1] = x[n] + 1/6 (k1 + 2 k2 + 2 k3 + k4);
y[n + 1] = y[n] + 1/6 (l1 + 2 l2 + 2 l3 + l4);
z[n + 1] = z[n] + 1/6 (m1 + 2 m2 + 2 m3 + m4);},
{n,0,tmax}];
sos = Table[{x[t], y[t], z[t]}, {t, 0, tmax}];
ListPointPlot3D[sos, ImageSize -> 300]
www.pdfgrip.com
p = Interpolation /@ Transpose@sos;
PlotPoints -> 100,
ColorFunction -> (Hue[#4] &), ImageSize -> 300]
Figures 6.6 and 6.7 show the Lorenz system without and with interpolation,
respectively.
Example 7 (Rössler Attractor) Solve the fractional Rössler8 attractor system, using
the RK4 method:
⎧
⎪
⎪ D 0.98 x(t) = −y − z, x(0) = 1,
⎪
⎪
⎪
⎨
D 0.98 y(t) = x + 0.2y, y(0) = 1,
⎪
⎪
⎪
⎪
⎪
⎩ 0.98
D z(t) = 0.2 + z(x − 8), z(0) = 1.
40
30
20 20
10
0
0
–10
10 –20
20
Fig. 6.6 Lorenz system without interpolation
8 O.E. Rössler(1940–).
www.pdfgrip.com
20
20
40
30
20
10
10
0
10
20
Fig. 6.7 Lorenz system with interpolation
The Mathematica solution is:
Clear["‘*"]
f[t_, x_, y_, z_] := -y - z;
g[t_, x_, y_, z_] := x + 0.2*y;
p[t_, x_, y_, z_] := 0.2 + z*(x - 8);
a = 0.998; d = 0.01; x[0] = 1;
y[0] = 1; z[0] = 1; tmax = 2000;
h = (d)â/Gamma[a + 1];
Do[{k1 = h f[t[n], x[n], y[n], z[n]];
l1 = h g[n*d, x[n], y[n], z[n]];
m1 = h p[n*d, x[n], y[n], z[n]];
k2 = h f[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
l2 = h g[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
m2 = h p[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
www.pdfgrip.com
k3 = h f[n*d + h/2, x[n] + k2/2, y[n] + l2/2,

z[n] + m2/2];
l3 = h g[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
m3 = h p[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
x[n + 1] = x[n] + 1/6 (k1 + 2 k2 + 2 k3 + k4);
y[n + 1] = y[n] + 1/6 (l1 + 2 l2 + 2 l3 + l4);
z[n + 1] = z[n] + 1/6 (m1 + 2 m2 + 2 m3 + m4);},
{n, 0, tmax}]
ListPointPlot3D[Table[{x[t], y[t], z[t]},
{t, 0, tmax}],
ImageSize -> 300]
Figure 6.8 shows the 3D Rössler attractor solution (x(t), y(t), z(t)).
0.10
10
0.05
5
0.00
5 0
5
5
Fig. 6.8 The 3D Rössler attractor solution (x(t), y(t), z(t))

www.pdfgrip.com
Example 8 Find the solution of the fractional Volta attractor:

⎧
⎪
⎪ D 0.998 x(t) = −x − 5y − yz, x(0) = 8,
⎪
⎪
⎪
⎨
D 0.998 y(t) = −85x − y − xz, y(0) = 2,
⎪
⎪
⎪
⎪
⎪
⎩ 0.998
D z(t) = 0.5z + xy + 1, z(0) = 1.
Solution in Mathematica, based on the RK4 method.
Clear["‘*"]
f[t_, x_, y_, z_] := -x - 5*y - z*y;
g[t_, x_, y_, z_] := - 85*x - y - x*z;
p[t_, x_, y_, z_] := 0.5*z + x*y + 1;
a = 0.998; d = 0.001; tmax = 10000;
x[0] = 8; y[0] = 2; z[0] = 1;
h = (d)â/Gamma[a + 1];
Do[{k1 = h f[t[n], x[n], y[n], z[n]];
l1 = h g[n*d, x[n], y[n], z[n]];
m1 = h p[n*d, x[n], y[n], z[n]];
k2 = h f[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
l2 = h g[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
m2 = h p[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
k3 = h f[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
l3 = h g[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
m3 = h p[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
x[n + 1] = x[n] + 1/6 (k1 + 2 k2 + 2 k3 + k4);
y[n + 1] = y[n] + 1/6 (l1 + 2 l2 + 2 l3 + l4);
z[n + 1] = z[n] + 1/6 (m1 + 2 m2 + 2 m3 + m4);},
{n,0,tmax}]
ListPointPlot3D[Table[{x[t], y[t], z[t]},{t,0,tmax}],
ImageSize -> 300]
sos = Table[{x[n], y[n], z[n]}, {n, 1, 10000}];
ParametricPlot3D[Evaluate@Through@p@t,{t,1,10000},
ImageSize->300]
www.pdfgrip.com
Figures 6.9 and 6.10 show the fractional Volta attractor solution without and with
interpolation, respectively.
Example 9 (Chua Attractor System) Find the numeric solution of the Chua9 attrac-
tor system:
⎧
⎪
⎪ D 0.998 x(t) = 40(y − x), x(0) = 0,
⎪
⎪
⎪
⎨
D 0.998 y(t) = (28 − 40)x + 28y − xz, y(0) = 1,
⎪
⎪
⎪
⎪
⎪
⎩ D 0.998 z(t) = xy − 2z, z(0) = 0.
Solution The solution in Mathematica is:
Clear["‘*"]
f[t_, x_, y_, z_] := 40*(y - x);
g[t_, x_, y_, z_] := x*(28 - 40) + 28*y - x*z;
50 40
100 20
0
100
50
20
0
50 40
100
Fig. 6.9 Fractional Volta attractor solution without interpolation
9 L.O. Chua (1936–).

www.pdfgrip.com
40 100
20
50
0
20 0
40 50
50
100
Fig. 6.10 Fractional Volta attractor solution with interpolation
p[t_, x_, y_, z_] := x*y - 2* z;

a = 0.998;
d = 0.01;
x[0] = 0;
y[0] = 1;
z[0] = 0;
tmax = 10000;
h = (d)â/Gamma[a + 1]; Do[{k1 = h f[t[n], x[n],
y[n], z[n]];
l1 = h g[n*d, x[n], y[n], z[n]];
m1 = h p[n*d, x[n], y[n], z[n]];
k2 = h f[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
l2 = h g[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
m2 = h p[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
k3 = h f[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
www.pdfgrip.com
l3 = h g[n*d + h/2, x[n] + k2/2, y[n] + l2/2,

z[n] + m2/2];
m3 = h p[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
x[n + 1] = x[n] + 1/6 (k1 + 2 k2 + 2 k3 +k4);
y[n + 1] = y[n] + 1/6 (l1 + 2 l2 + 2 l3 +l4);
z[n + 1] = z[n] + 1/6 (m1 + 2 m2 + 2 m3 +m4);},
{n,0,tmax}];
ListPointPlot3D[sos, ImageSize -> 300]
PlotPoints -> 100,
Figures 6.11 and 6.12 show the fractional Chua attractor solution without and
with interpolation, respectively.
40
30
20 20
10
0 10
0
10
0
10
10
20 20
Fig. 6.11 The Chua attractor solution without interpolation

www.pdfgrip.com
20
10
0
–10
–20
40
30
20
10
–10
0
10
20
Fig. 6.12 The Chua attractor solution with interpolation
6.5.3 A More General System
Let the following general system of FDE, with initial conditions:

⎧ α
⎨ D x(t) = f (t, x(t), y(t)), x(t0 ) = x0 ,
⎩
D β y(t) = g(t, x(t), y(t)), y(t0 ) = y0 ,
where we suppose that 0 < β ≤ α ≤ 1.

We will establish here a RK2 approximate solution for this system.
We will denote the temporal step with h, so that the discrete values of time are:
tn = t0 + nh, n ∈ N.
The approximate solution of second order of this system can be written as:
hα
xn+1 = xn + (K1 + K2 ),
(α + 1)
www.pdfgrip.com
hβ
yn+1 = yn + (L1 + L2 ),
(β + 1)
where we used the notations:
K1 = f (tn , xn , yn ), L1 = g(tn , xn , yn ),

hα hα K1 hβ L1
K2 = f tn + , xn + , yn + ,
(α + 1) (α + 1) 2 (β + 1) 2

hβ hα K1 hβ L1
L2 = g t n + , xn + , yn + .
(β + 1) (α + 1) 2 (β + 1) 2
Proof In this proof the calculations will be presented schematically.

By analogy with Subsection 4.11.1, we will establish an approximation using the
Taylor expansion:
tα tα
xn+1 = xn + c1 K1 + c2 K2
(α + 1) (α + 1)
tα tα
yn+1 = yn + d1 L1 + d2 L2
(α + 1) (α + 1)
where
K1 = f (tn , xn , yn ),
L1 = g(tn , xn , yn ),

hα hα hβ
K2 = f tn + af α , xn + K1 bf α , yn + L1 bfβ ,
(α + 1) (α + 1) (β + 1)

hβ hα hβ
L2 = g t n + agβ , xn + K1 bgα , yn + L1 bgβ ,
(β + 1) (α + 1) (β + 1)
where the constants c1 , c2 , d1 , d2 , af α , bf α , agα , bgα , afβ , bfβ , agβ , and bgβ will
be established.
We define the errors E(h) and F (h) of expansions
hα h2α
E(h) = x(tn + h) − x(tn ) = D α x(tn ) + D 2α x(tn ) + O(h3α ),
(α + 1) (2α + 1)
hβ h2β
F (h) = y(tn + h) − y(tn ) = D β y(tn ) + D 2β y(tn ) + O(h3β ),
(β + 1) (2β + 1)
www.pdfgrip.com
and the derivatives:
D α E(h) = f (tn , xn , yn ) + O(hα ) ⇒ D α E(0) = f (tn , xn , yn ),
D β F (h) = g(tn , xn , yn ) + O(hβ ) ⇒ D β E(0) = g(tn , xn , yn ).
On the other hand
D α E(h) = c1 K1 + c2 K2 + O(hα) ⇒ D α E(0) = (c1 + c2 )f (tn , xn , yn ),
D β E(h) = d1 L1 + d2 L2 + O(hα) ⇒ D β E(0) = (d1 + d2 )g(tn , xn , yn ),
hence, by minimization (h → 0) we obtain:

⎧
⎨ c1 + c2 = 1,
⎩
d1 + d2 = 1.
The derivative of E(h) is
D 2α E(h) = D α (D α E(h))

hα
= D (c1 K1 + c2 K2 ) + D c2
α α
D K2 + O(hα ),
α
(α + 1)
and for h → 0 we have
D 2α E(0) = 2c2 D 2α K2 (0).
But:
D α f (t, x(t), y(t)) = ft + fx D α x + fy D α y = ft + fx f,
Dα y = 0 because: α > β,
we obtain
⎧
⎪
⎪ 2c2 af α = 1,
⎪
⎪
⎪
⎨
2c2 bf α1 = 1,
⎪
⎪
⎪
⎪
⎪
⎩ 2c b
2 f α2 = 1.
The calculations will continue with the derivative. Similar computations can be
carried out in the case of RK4 method.
www.pdfgrip.com
It results:
1 hα
xn+1 = xn + (K1 + 2K2 + 2K3 + K4 ),
6 (α + 1)
1 hβ
yn+1 = yn + (L1 + 2L2 + 2L3 + L4 ).
6 (β + 1)
Example 1 (Modified Duffing System) Find the RK4 solution for the modified
fractional Duffing system:
⎧ 0.998
⎨D x(t) = y, x(0) = 1,
⎩
D 0.50 y(t) = −x + 0.25(1 − x 3 )y, y(0) = 0.
The RK4 solution in Mathematica is:
Clear["‘*"];
a = 0.998; b = 0.5; x[0] = 1; y[0] = 0;
h = (0.2)â/Gamma[a + 1]; l = (0.2)^b/Gamma[b + 1];
n = 10000;
Do[t[i] = 0.0 + (0.2)*i, {i, 0, n}]
f[t_, x_, y_] := y
g[t_, x_, y_] := -x + 0.25*(1 - x^2)*y
Do[{K1 = f[t[i], x[i], y[i]], L1 = g[t[i], x[i],
y[i]],
K2 = f[t[i] + h, x[i] + h*K1, y[i] + h*L1],
L2 = g[t[i] + l, x[i] + l*K1, y[i] + l*L1],
K3 = f[t[i] + h/2, x[i] + h*K2/2, y[i] + h*L2/2],
L3 = g[t[i] + l/2, x[i] + l*K2/2, y[i] + l*L2/2],
K4 = f[t[i] + h, x[i] + h*K3, y[i] + h*L3],
L4 = g[t[i] + l, x[i] + l*K3, y[i] + l*L3],
x[i + 1] = x[i] + (K1 + 2 K2 + 2 K3 + K4)*h/6,
y[i + 1] = y[i] + (L1 + 2 L2 + 2 L3 + L4)*l/6},
{i,0,n}];
ListPlot[Table[{x[n], y[n]}, {n, 0, 10000}],
Frame -> True]
sol = Table[{x[n], y[n]}, {n, 1, 1000}];
ParametricPlot[Evaluate@Through@p@t, {t, 1, 1000},
Frame -> True]
www.pdfgrip.com
Fig. 6.13 RK4 solution

(x(t), y(t)) of the modified 1.5
fractional Duffing system
without iteration
1.0
0.5
0.0
–0.5
–1.0
–1.5
–0.5 0.0 0.5 1.0
Fig. 6.14 RK4 solution

1.5
(x(t), y(t)) of the modified
fractional Duffing system
with iteration
1.0
0.5
0.0
–0.5
–1.0
–1.5
–0.5 0.0 0.5 1.0
Figures 6.13 and 6.14 show the RK4 solution (x(t), y(t)) of the modified
fractional Duffing system without and with iteration, respectively.
Remark Starting from the fractional RK procedures introduced in this section, the
reader can establish similar procedures for the other RK methods, established for
the integer order cases [1].
www.pdfgrip.com
The Fractional Colpitts Oscillator

Solve the fractional Colpitts10 oscillator described by the system of FDE:
⎧
⎪
⎪ D 0.998 x(t) = y, x(0) = 0,
⎪
⎪
⎪
⎨
D 0.998 y(t) = z, y(0) = 0.4,
⎪
⎪
⎪
⎪
⎪
⎩ 0.998
D z(t) = −z − x − 10−9 (exp(y) − 1), z(0) = 0,
using RK4 method in Mathematica.

Solution
Clear["‘*"]
f[t_, x_, y_, z_] := y;
g[t_, x_, y_, z_] := z;
p[t_, x_, y_, z_] := -z - x - 10^(-9) (Exp[y] - 1);
a = 0.998;
d = 0.01;
x[0] = 0;
y[0] = 0.4;
z[0] = 0;
tmax = 10000;
h = (d)â/Gamma[a + 1];
Do[{k1 = h f[t[n], x[n], y[n], z[n]];
l1 = h g[n*d, x[n], y[n], z[n]];
m1 = h p[n*d, x[n], y[n], z[n]];
k2 = h f[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
l2 = h g[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
m2 = h p[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
k3 = h f[n*d + h/2, x[n] + k2/2, y[n] + l2/2,

z[n] + m2/2];
l3 = h g[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
m3 = h p[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
x[n + 1] = x[n] + 1/6 (k1 + 2 k2 + 2 k3 + k4);
10 E.H. Colpitts (1872–1949).

www.pdfgrip.com
y[n + 1] = y[n] + 1/6 (l1 + 2 l2 + 2 l3 + l4);

z[n + 1] = z[n] + 1/6 (m1 + 2 m2 + 2 m3 + m4);},
{n, 0, tmax}]
{t, 0, tmax}], ImageSize -> 300]
PlotPoints -> 100,
Figures 6.15 and 6.16 show the RK4 solution (x(t), y(t)) of the fractional
Colpitts oscillator system without and with iteration, respectively.
The Fractional Sprott Oscillator
Solve the fractional Sprott11 oscillator described by the system of FDE:
⎧
⎪
⎪ D 0.998 x(t) = y, x(0) = −0.5,
⎪
⎪
⎪
⎨
D 0.998 y(t) = z, y(0) = 0,
⎪
⎪
⎪
⎪
⎪
⎩ 0.998
D z(t) = −z − x − 10−9 (exp(y) − 1), z(0) = 0,
using RK4 method in Mathematica.
Fig. 6.15 RK4 solution of

the fractional Colpitts
oscillator system without
iteration
50
0 20
–50 0
–50 –20
0
–40
11 D.A. Sprott (1930–2013).

www.pdfgrip.com
20 –50
0
–20 0
–40
50
–50
Fig. 6.16 RK4 solution of the fractional Colpitts oscillator system with iteration
Solution
Clear["‘*"]
f[t_, x_, y_, z_] := y;
g[t_, x_, y_, z_] := z;
p[t_, x_, y_, z_] := -x - y - Sign[1 + 4*y];
a = 0.998; d = 0.01; x[0] = -0.5; y[0] = 0;
z[0] = 0; tmax = 2000; h = (d)â/Gamma[a + 1];
Do[{k1 = h f[t[n], x[n], y[n], z[n]];
l1 = h g[n*d, x[n], y[n], z[n]];
m1 = h p[n*d, x[n], y[n], z[n]];
k2 = h f[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
l2 = h g[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
m2 = h p[n*d + h/2, x[n] + k1/2, y[n] + l1/2,
z[n] + m1/2];
www.pdfgrip.com
k3 = h f[n*d + h/2, x[n] + k2/2, y[n] + l2/2,

z[n] + m2/2];
l3 = h g[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
m3 = h p[n*d + h/2, x[n] + k2/2, y[n] + l2/2,
z[n] + m2/2];
x[n + 1] = x[n] + 1/6 (k1 + 2 k2 + 2 k3 + k4);
y[n + 1] = y[n] + 1/6 (l1 + 2 l2 + 2 l3 + l4);
z[n + 1] = z[n] + 1/6 (m1 + 2 m2 + 2 m3 + m4);},
{n, 0, tmax}]
{t, 0, tmax}], ImageSize -> 300]
p = Interpolation /@ Transpose@sos ;
PlotPoints -> 100,
Figures 6.17 and 6.18 show the RK4 solution of the fractional Sprott oscillator
system without and with iteration, respectively.
Fig. 6.17 RK4 solution of

the Sprott oscillator system
without iteration
0 0
5
www.pdfgrip.com
–2 –1
2 0 1
0
–2
–4
5
Fig. 6.18 RK4 solution of the Sprott oscillator system with iteration
6.5.4 A Vectorial Runge–Kutta Algorithm
We illustrate the vectorial Runge–Kutta algorithm by means of four examples:

Example 1 (Van Der Pol Fractional Equation)
⎧ α
⎨ D x1 (t) = x2 , x1 (0) = 0,
⎩
D α x2 (t) = −x1 + 2 (1 − x12 ) x2 , x2 (0) = 1,
for α = 0.998.
Solution In Mathematica the solution will be:
Clear["‘*"]
a = 0.998
RK4[f_, x0_, A_, B_, n_] :=
www.pdfgrip.com
Module[{h, K1, K2, K3, K4, Sol = {{A, x0}},

x = x0, t}, h = (B - A)/n;
Do[t = A + k hâ/Gamma[a + 1];
K1 = hâ/Gamma[a + 1] f[t, x];
K2 = hâ/Gamma[a + 1] f[t + (1/2) hâ/Gamma[a + 1],
x + (1/2) K1];
x + (1/2) K2];
K4 = hâ/Gamma[a + 1] f[t + hâ/Gamma[a + 1],
x + K3];
x = x + (1/6) K1 + (1/3) K2 + (1/3) K3 + (1/6) K4;
Sol = Append[Sol, {t, x}], {k, 1, n}];
Sol]
F[t_, x_] := { x[[2]], 2 (1 - x[[1]] ^2) x[[2]]
- x[[1]]};
Solution = RK4[F, {0, 1}, 0.0, 100.0, 5000];

A = 0
ListPlot[Take[Solution[[All, 2]], 5000],
ImageSize -> 300,
PlotStyle -> {Blue}]
Figure 6.19 shows the (x1 (t), x2 (t)) vector solution of the Van der Pol system.
Fig. 6.19 The (x1 (t), x2 (t)) 4

vector solution of the Van der
Pol system
–2 –1 1 2
–2
–4
www.pdfgrip.com
Example 2 Solve the fractional Rössler attractor system, using the vector RK4
method:
⎧
⎪
⎪ D 0.98 x1 (t) = −x2 − x3 , x1 (0) = 0,
⎪
⎪
⎪
⎨
D 0.98 x2 (t) = x1 + 0.2x2 , x2 (0) = 1,
⎪
⎪
⎪
⎪
⎪
⎩ 0.98
D x3 (t) = 0.2 + x3 (x1 − 8), x3 (0) = 0.
Solution
Clear["‘*"]
a = 0.998
RK4[f_, x0_, A_, B_, n_] :=
x = x0, t}, h = (B - A)/n;
K2 = hâ/Gamma[a + 1] f[t +
(1/2) hâ/Gamma[a + 1], x + (1/2) K1];
(1/2) hâ/Gamma[a + 1], x + (1/2) K2];
hâ/Gamma[a + 1], x + K3];
x = x + (1/6) K1 + (1/3) K2 +
(1/3) K3 + (1/6) K4;
Sol]
F[t_, x_] := {-x[[2]] - x[[3]], x[[1]]
+ 0.2 x[[2]], 0.2 + x[[1]] x[[3]] -
5.7 x[[3]]};
Solution = RK4[F, {0, 1.0, 0}, 0.0, 200.0, 5000];

A = 0
ListPointPlot3D[Take[Solution[[All, 2]], 1000],
Figure 6.20 shows the (x1 (t), x2 (t)) vector solution of the Rössler attractor.
Example 3 (Volta Fractional Attractor) Find the solution of the fractional Volta
attractor using the vector RK4 algorithm:
⎧
⎪
⎪ D 0.998 x1 (t) = −x1 − 5x2 − x2 x3 , x1 (0) = 8,
⎪
⎪
⎪
⎨
D 0.998 x2 (t) = −85x1 − x2 − x1 x3 , x2 (0) = 2,
⎪
⎪
⎪
⎪
⎪
⎩ D 0.998 x (t) = 0.5x + x x + 1,
3 3 1 2 x3 (0) = 1.
www.pdfgrip.com
0.4
0.3
0.2 5
0.1
0.0 0
–5
0 –5
10 –10
Fig. 6.20 The (x1 (t), x2 (t), x3 (t)) vector solution of the Rössler attractor
Solution
In Mathematica can be written:
Clear["‘*"]
a = 0.998
RK4[f_, x0_, A_, B_, n_] :=
x = x0, t}, h = (B - A)/n;
x + (1/2) K1];
x + (1/2) K2];
x + K3];
x = x + (1/6) K1 + (1/3) K2 + (1/3) K3 + (1/6) K4;
Sol]
www.pdfgrip.com
F[t_, x_] := {-x[[1]] - 5 x[[2]] - x[[2]] x[[3]],

-85 x[[1]]
- x[[2]] - x[[1]] x[[3]], x[[1]] x[[2]] + 0.5
x[[3]] + 1};
Solution = RK4[F, {8.0, 2.0, 1.0}, 0.0, 100.0, 5000];

A = 0
Figure 6.21 shows the (x1 (t), x2 (t)) vector solution of the Volta attractor.
Example 4 (Lorenz Fractional Attractor) Solve the Lorentz attractor system:
⎧
⎪
⎪ D 0.98 x1 (t) = −10(x1 − x2 ), x1 (0) = 1,
⎪
⎪
⎪
⎨
D 0.98 x2 (t) = 28x1 − x2 − x1 x3 , x2 (0) = 1,
⎪
⎪
⎪
⎪
⎪
⎩ D 0.98 x (t) = x x − 8 x ,
3 1 2 3 3 x3 (0) = 1.
using the vector RK4 method, in Mathematica.
40
–50
20
–100
0
0
20 –20
40
–40
60
Fig. 6.21 The (x1 (t), x2 (t), x3 (t)) vector solution of the Volta attractor
www.pdfgrip.com
40
30
20 20
10
0
0
–10
10
–20
20
Fig. 6.22 The (x1 (t), x2 (t), x3 (t)) vector solution of the Lorenz attractor
Solution
Clear["‘*"]
a = 0.998
RK4[f_, x0_, A_, B_, n_] :=
x = x0, t}, h = (B - A)/n;
x + (1/2) K1];
x + (1/2) K2];
x + K3];
x = x + (1/6) K1 + (1/3) K2 + (1/3) K3 + (1/6) K4;
Sol]
F[t_, x_] := {10 (x[[2]] - x[[1]]), x[[1]]
(28 - x[[3]])
www.pdfgrip.com
References 185
- x[[2]], x[[1]] x[[2]] - 8/3 x[[3]]};
Solution = RK4[F, {1.0, 1.0, 1.0}, 0.0, 100.0, 5000];

A = 0
Figure 6.22 shows the (x1 (t), x2 (t)) vector solution of the Lorenz attractor.
References
1. Butcher, J. C. (1987). The numerical analysis of ordinary differential equations, Runge-Kutta

and general linear methods. New York: Wiley-Interscience.
2. Drăgănescu, G. E. (2006). Application of a variational iteration method to linear and nonlinear
viscoelastic models with fractional derivatives. Journal of Mathematical Physics, 47, 082902.
3. He, J.-H. (1999). Variational iteration method - a kind of non-linear analytical technique: Some
examples. International Journal of Non-linear Mechanics, 34, 699–708.
4. He, J.-H. (2006). Some asymptotic methods for strongly nonlinear equations. International
Journal of Modern Physics B, 20, 1141–1199.
5. Wu, G.-C. (2011). A fractional variational iteration method for solving fractional nonlinear
differential equations. Computers & Mathematics with Applications, 61, 2186–2190.
6. Wu, G.-C., & Baleanu, D. (2013). Variational iteration method for fractional calculus - a
universal Laplace transform. Advances in Difference Equations, 18, 1–9.
www.pdfgrip.com
Index
B F
Bernoulli numbers, 129 Fourth order Runge–Kutta method, 153
Bessel function of first kind, 115 Fractional calculus, v
Bessel generalized fractional differential Fractional Colpitts oscillator, 175
equation, 113 Fractional Cosine, 87
Beta function, 7 Fractional derivative, v
Fractional differential equation, vi, 47
Fractional Duffing system, 159
Fractional Rössler attractor system, 181
C
Fractional Rössler system, 163
Caputo, 24
Fractional Sine, 87
Caputo FDE, 47
Fractional Sprott oscillator, 176
Caputo fractional derivative, 24, 44
Fractional Van der Pol system, 157
Caputo fractional integral, 43
Fractional Volta attractor, 166, 181
Cauchy formula, 23
Chebyshev norm, 48
Chua attractor, 167 G
Complementary error function, 11 Galerkin method, 135
Cornu fractional differential system, 88 Gamma function, 1
Cornu Fractional System, 87 Gauss’ formula, 5
Gauss multiplication formula, 11
Generalized MacLaurin series, 109
D Ghelfand fractional differential equation, 73
Dirichlet equality, 21 Gram determinant, 96
Dirichlet theorem, 18
H
Hermite generalized fractional differential
E equation, 110
E function, 13
Error function, 11
Euler integral, 1 I
Euler method, 143 Imaginary error function, 11
Euler’s constant, 6 Inverse Laplace transform, 33
Exponential integral function, 11 Iterative Euler formula, 143

Systems and Complexity 25, https://doi.org/10.1007/978-3-030-00895-6
www.pdfgrip.com
188 Index
L R
Lagrange multiplier, 122 Residues, 35
Lane and Emden fractional differential Riemann, 17
equation, 104 Riemann–Liouville derivative, 17
Laplace transform, 33 Riemann–Liouville FDE, 47
Least squares method, 124 Riemann–Liouville fractional derivative, 43
Left fractional integral, 17 Riemann–Liouville fractional integral, 43
Legendre duplication, 10 Right fractional integral, 17
Legendre fractional differential equation, 111 Runge–Kutta method, 150, 153
Liouville, 17
Lorenz attractor, 148, 162
Lorenz attractor system, 183 S
Lorenz fractional system, 118 Second order Runge–Kutta algorithm, 150
Lotka system, 116 Stirling formula, 37
T
M Taylor, B., 25
Method of Picard, 52 Taylor generalized formula, 110
Mittag-Leffler, 11 Two parameter Mittag-Leffler function, 11
Modified fractional Duffing system, 173
V
Van Der Pol fractional equation, 179
O Variational iteration method, 121
One parameter Mittag-Leffler function, 11 Vectorial Runge–Kutta algorithm, 179
Voltera integral, 48
P W
Post formula, 36 Weierstrass form, 6
Power series method, 94 Weierstrass test, 48, 51

Introduction To Fractional Differential Equations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Fractional Differential Equations

Uploaded by

Copyright:

Available Formats

Nonlinear Systems and Complexity

Series Editor: Albert C. J. Luo

Nonlinear Systems and Complexity

Nonlinear Systems and Complexity provides a place to systematically summarize

More information about this series at http://www.springer.com/series/11433

Constantin Milici • Gheorghe Drăgănescu

Constantin Milici Gheorghe Drăgănescu

ISSN 2195-9994 ISSN 2196-0003 (electronic)

Library of Congress Control Number: 2018956771

© Springer Nature Switzerland AG 2019

The concept of fractional derivative (Fractional derivative (FD)) was introduced

introduced a series of identification methods for fractional dynamic systems [13].

Moreover, a generalized method, based on decomposition and LT, is presented [23–

Timişoara, Romania Constantin Milici

10. Drăgănescu, G. E. (2006). Application of a variational iteration method to

4 Fractional Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

1.1 Euler’s Function

1.1.1 Gamma Function

We start by considering the Gamma function, or second order Euler1 integral,

Theorem Function (p) is convergent for p > 0.

© Springer Nature Switzerland AG 2019 1

Fig. 1.1 The plot of

Since e−x is decreasing on the interval [0, 1], from x = 0, we have:

The integral (1.1) is convergent for p > 0 and divergent for p ≤ 0.

(p + 1) = p (p). (1.2)

1.1 Euler’s Function 3

(p + n) = (p + n − 1) . . . (p + 1) p (p), (1.3)

Using the substitution p + n = α, it results after calculations:

6. Using the identity (1.2) we obtain:

7. We can prove the identity [4]:

1.1 Euler’s Function 5

10. The Gauss’2 formula is:

Proof We express e−x as:

For x = t k ⇒ dx = k dt, resulting:

Integrating by parts we obtain:

Repeating this operation it follows:

2 J.C.F. Gauss (1777–1855).

excepting the values Re(p) = 0, −1, −2, . . ..

Proof The Gauss’ formula can be written as:

excepting for Re(p) = 0, −1, −2, . . .. We can write also:

k p = ep ln k = ep(ln k − 1 − 1/2 − . . . − 1/k) ep + p/2 + . . . + p/k ,

ep(ln k − 1 − 1/2 − . . . − 1/k) ep + p/2 + . . . + p/k

12. Reflection property of the Gamma function is given by:

3 K.T.W. Weierstrass (1815–1897).

1.1 Euler’s Function 7

Proof From the Weierstrass’ formula we have:

Using the changes of variables

1.1.2 Beta Function

Here we consider the Beta function, denoted Beta function (B).

where Re(p) > 0 and Re(q) > 0.

Fig. 1.2 The plot of

B(p, q) = B(q, p).

x p (1 − x)q−2 = x p−1 (1 − x)q−2 − x p−1 (1 − x)q−1 ,

1.1 Euler’s Function 9

3. For every p > 0 and q > 0, it is valid the identity:

Proof The product (p)(q) can be written as:

We use the notation t + s = x, for 0 < t < ∞ and 0 < t < ∞.

(p)(q) = (p + q)B(p, q).

For any natural numbers m, n we obtain:

5. Legendre4 duplication formula:

Proof Substituting p = q in B(p, q):

4 A.M. Legendre (1752–1833).

1.3 Mittag-Leffler Function 11